**E**ff**ects of Soccer Training on Body Balance in Young Female Athletes Assessed Using Computerized Dynamic Posturography**

#### **Gra ˙zyna Olchowik and Agata Czwalik \***

Department of Biophysics, Medical University of Lublin, K. Jaczewskiego 4, 20-090 Lublin, Poland; grazyna.olchowik@umlub.pl

**\*** Correspondence: agata.czwalik@umlub.pl; Tel.: +48-81-448-6330

Received: 18 December 2019; Accepted: 23 January 2020; Published: 3 February 2020

**Abstract:** The aim of this study was to determine the effect of regular soccer training on the balance system for young women. Computerized dynamic posturography of female footballers (n = 25) and control group (n = 50) was assessed during three tests: Sensory Organization Test, Motor Control Test, and Adaptation Test. Statistically significant differences between the groups was found in Composite Equilibrium Score with higher values, indicating better postural stability, for footballers. Regular trainees also showed better usefulness of vestibular system while maintaining balance. Weight symmetry of the lower limbs during Motor Control Test also showed statistically significant differences between the groups. This study shows that female footballers have better postural stability than their inactive peers and that regular workouts may improve the balance system.

**Keywords:** posture stability; balance; football; exercise; training

#### **1. Introduction**

Football is the most popular sport discipline with around 200,000 professional players and 240 million amateur players. It is responsible for almost 10% of sports injuries requiring medical attention in adolescents [1,2]. This discipline requires players to have unprecedented coordination to cope with rapidly changing external conditions. With regular practice a player acquires precision in movement and more muscle mass. Frequent training helps the player acquire the ability to execute appropriate strategic movements to effectively target his/her opponent's goal and to prevent serious injuries. To evaluate motor coordination a test was adopted many years ago to measure maximum rotation to the left and right when jumping in the air with both feet [3]. In modern training it is necessary to constantly monitor the sportsperson's coordination and balance [4,5]. Presently there are many different methods to assess training progress, physical activity, and body postural stability, but the most important conditions in which football players should be assessed are dynamic conditions. Postural stability in athletes has been reported widely in several sports disciplines [6,7]. That suggest that the type of sport and repetitive training may affect the balance system control. Some researchers found that football players were better that other athletes [6,8] and that the level of playing experience influences postural control performance and adopted motor strategies [9].

The human body's static and dynamic balance is maintained by the posture control system, which coordinates the stimuli received by the vestibular system, the visual system, and proprioceptors, and also provides a selection of optimal postural responses aimed at preventing falls [10,11]. The posture control system tracks the position of the center of gravity (COG) over an area defined by the outline of the human feet, thus guaranteeing a stable posture [12]. The body's multi-segment construction, the height of the COG above the base of support (BOS), and a relatively small BOS, results in the

human body being unstable when in an upright position. The balance control system therefore needs to constantly analyze and counterbalance all the destabilizing factors through proper stimulation of the relevant muscle groups [12,13]. The central nervous system watches over the appropriate choice of reaction and the stimulation of the appropriate muscle reflexes. To correct posture, it adapts a movement strategy based on its analysis of linear and angular accelerations of individual body segments. Stability control of body posture depends on many factors, which includes a range of movements in the joints, muscle strength, speed and precision of movement, and the ability to perceive body positioning in space ("body sense") [14].

Computerized Dynamic Posturography (CDP) allows the individual components of the human balance system to be evaluated. During CDP, signals from appropriate senses involved in maintaining balance are evaluated. The postural response time and the reaction time to unexpected support platform changes, with an appropriate motor response, is also determined as well as the efficiency of adaptive mechanisms [15]. CDP is the gold standard to differentiate between sensory, motor, and central adaptive impairments to postural control [16]. In football it can be an opportunity to track the rehabilitation of postural control impairment after anterior cruciate ligament injury, which is one of the most common in this sport [17].

The aim of this study was to compare the behavior of the balance system between two groups of young women and to determine the effect of regular practice. One group consisted of young female footballers who train regularly and the other, a group of peers who do not participate in any regular sporting activity. We hypothesize that football players would show better postural balance performance.

#### **2. Materials and Methods**

#### *2.1. Participants*

The study group consisted of 25 young (age = 18.9 ± 4.5 year) female footballers from AZS-PSW Biała Podlaska, a club in the Polish Women's Football League. Each footballer in this study group has been attending training sessions for at least 6 years with five training sessions per week of 90 min duration. The control group consisted of 50 students (age = 20.7 ± 1.2 year) from the Medical University of Lublin, who had declared a lack of sporting activity. Both groups were chosen such that there were no statistically significant differences between them in height, weight, or body mass index (BMI) (Table 1). All subjects participating in the study had the same functional preferences (right-sided handedness, footedness, and eyedness) and same postural lateral preferences (hand-clasping, arm-folding, leg-crossing, and stair climbing) [18].



M = mean, SD—standard deviation, *p*—probability value.

#### *2.2. Procedure*

The study was conducted with the approval of the Bioethics Committee at the Medical University of Lublin (KE 0254/195/2011). At the beginning, participants were informed about the purpose of the study and asked to complete a questionnaire and to sign the consent for the study. The questionnaire included questions about general health: all diseases or health problems, surgeries or hospitalizations, brain injuries, loss of consciousness episodes, bone/muscle/joint injuries as well as any medication or

drugs (including alcohol) taken in the past month. Another part of the questionnaire was about physical activity: what sport or physical activity the participant undertakes, for how long, and how many times per week. Afterwards each participant was accurately weighed and measured, and a short lateralization preference test was performed (e.g., which hand the participant uses for drawing/throwing a small object, which leg for kicking a football or stepping onto a chair, and which eye for looking into a bottle or a door viewer). The final stage of the study was a posturographic examination during which the participant stood barefoot on the posturographic platform and was secured with a special harness.

#### *2.3. Instruments*

Posturographic tests were performed using an Equitest posturograph manufactured by NeuroCom International®. The device consists of a dynamic force plate, visual surround, and a computer with software. Both the force plate and visual surround are moveable (±10◦ rotation for both and a maximum angular velocity of 50◦/s and 15◦/s, respectively). The study protocol included the following tests: Sensory Organization Test, Motor Control Test, and the Adaptation Test.

The Sensory Organization Test (SOT) evaluates the usefulness of signals coming from the different senses involved in maintaining body balance. This test is performed using six sensory stimulation conditions, during which visual stimuli are changed and a rotation of the foot support platform, or movements of the visual surround, are introduced. During the first three tests (SOT1–SOT3) the foot support platform is stationary while the visual information is varied: SOT1—eyes open, SOT2—eyes closed, SOT3—moving visual surround. During these tests, analyzing postural stability determines the usefulness of the visual signal and the patient's ability to suppress visual stimuli which are contrary to reality. The next three tests (SOT4–SOT6) are performed with a moving foot support platform, which interferes with the information received by the proprioceptive system. As with the previous three tests, visual information is varied: SOT4—eyes open, SOT5—eyes closed, SOT6—moving visual surround. Two final trials evaluate the usefulness of the vestibular system, whose role increases significantly in the case of incorrect or missing stimuli from the remaining systems involved in posture control. During SOT, the Equilibrium Score (ES) that quantifies the COG sway or postural stability under each of the 3 trials of the 6 sensory conditions is evaluated. A score of 100 represents perfect balance (no sway) and a score of 0 indicates a fall. During SOT, Composite Equilibrium Score (CES)—a weighted average of all 6 individual scores ES—the body's COG displacement in the anterior–posterior direction, as well as the motor strategy (MS)—correctness of the selected postural strategy—is also assessed. A score near 100 indicates a full ankle strategy while a score near 0 indicates a full hip strategy with maximum shear force. In addition the Sensory Analysis (SRS) that determines a patient's ability to use input from the somatosensory (SOM), visual (VIS), vestibular (VES) system to maintain balance as well as the degree to which the patient relies on visual information whether it is correct or not (PREF) is assessed [19,20].

The Motor Control Test (MCT) is performed for 6 conditions using the foot support platform capable of forward and backward movements through small, medium, and large displacements. A series of three trials is performed for each condition during which the patient's ability to perform corrective movements is evaluated in response to unexpected changes in the foot support platform. The MCT analysis the Latency Response (L)—the time between start of platform movement and start of postural response—the Amplitude (A) of the postural response and the Weight Symmetry (WS)—a nondimensional quantity with a score of 100 indicating that weight is borne equally by the two legs. The WS score decreases to zero or increases to 200 when all the weight is borne by the left or right leg, respectively.

In the Adaptation Test (ADT), the patient is subjected to two series of sudden platform perturbations, one of which causes dorsiflexion (ATU) while the other causes plantarflexion (ATD) in the ankles. Each series consists of five trials. During subsequent trials, the patient should maintain a vertical posture, minimizing with every subsequent trial the amount of energy required to rebalance the body. This sway energy (SE) is determined after each platform perturbation and indicates the amount of COG displacement during each trial.

#### *2.4. Data Analysis*

Statistical analysis was performed using the STATISTICA 10 (StatSoft) computer program. To verify the normality of the data the Shapiro–Wilk test was used. Because normality tests failed to confirm that all the parameters variables studied had normal distribution (most probable reason for not normal distribution in footballers group was the small number of participants), all variables were analyzed with non-parametric statistics—the Mann–Whitney U-test—for differences between groups. Statistically significant changes were those with a statistical significance level of *p* < 0.05.

#### **3. Results**

The SOT results are shown in Figures 1–3. ES analysis (Figure 1) shows statistical significance (*p* < 0.05) for the CES, with higher values for footballers. Of the six SOT conditions, COG displacement differed significantly between the groups only for conditions ES5 and ES6 which provided conflicting information to the sense organs, carried out on a moving platform with eyes closed (ES5) or a moving visual surround (ES6). These results reflect significant differences in the use of vestibular stimuli (VES) between people training regularly and those not performing any regular sport.

**Figure 1.** Sensory Organization Test—Composite Equilibrium Score (CES) and Equilibrium Score (ES) results (where the ES suffix refers to a particular SOT condition) for footballers and control group. The symbol \* refers to a significant difference at *p* < 0.05.

**Figure 2.** Sensory Organization Test—sensory analysis (SRS) results for the somatosensory system (SOM), the visual system (VIS), the vestibular system (VES), and visual preference (PREF) for footballers and control group. The symbol \* refers to a significant difference at *p* < 0.05.

**Figure 3.** Sensory Organization Test—motor strategy (MS) for particular SOT conditions for footballers and control group.

The usefulness of the signals from the other sensory organs involved in the control of body balance (VIS, SOM, and PREF) did not highlight any significant differences between the two groups (Figure 2).

There was no impact on the selection of an appropriate motor strategy (MS1–MS6) by people playing football (Figure 3).

During MCT, statistically significant differences between the two groups were found in the symmetry of loading of the lower limbs during all the trials (Figure 4). Football players were characterized by disproportionate distribution of body weight with a predominance of left leg.

**Figure 4.** Motor Control Test—weight symmetry (WS) results for the lower limbs for small, medium and large platform transitions in backward (SB, MB, LB) and forward (SF, MF, LF) direction for footballers and control group. The symbol \* refers to a significant difference at *p* < 0.05.

Postural response latencies (Figure 5) and their amplitudes (Figure 6) did not reveal any statistically significant differences between the two groups, which means that neither the reaction time nor the angular velocity of COG during the trials depends on the physical activity.

**Figure 5.** Motor Control Test—postural latencies for left (LL) and right (LR) lower limb for small, medium and large platform transitions in backward (SB, MB, LB) and forward (SF, MF, LF) direction for footballers and control group.

**Figure 6.** Motor Control Test—amplitudes of the postural responses for left (AL) and right (AR) lower limb for small, medium and large platform transitions in backward (SB, MB, LB) and forward (SF, MF, LF) direction for footballers and control group.

In the ADT there were no statistically significant differences between the measured parameters for both groups, which means that the adaptive postural response system is independent of physical activity. The results are shown in Figure 7.

**Figure 7.** Adaptation Test—sway energy for 5 dorsiflexion (ATU) and 5 plantarflexion (ATD) condition for footballers and control group.

#### **4. Discussion**

Football is a sport discipline which, in addition to unprecedented coordination, requires exceptionally divided attention. The players must focus not only on the ball, but must also control their positions, the actions of the players in their team as well as those in the opposing team. They must analyze and foresee subsequent situations.

This study shows that professional footballers achieve significantly better results in the CES. The smaller COG displacements observed during all SOTs provides the evidence for the higher CES value for footballers who train regularly. The footballers also showed smaller COG displacements in the anterior–posterior direction for conditions which were inconsistent with the real signals from the somatosensory and visual systems. The footballers made better use of the signals from the vestibular system, which is the most reliable source of information regarding the positioning of the body's COG [15]. An analysis of the symmetry of the distribution of body weight gives a higher load asymmetry in the lower limbs in the footballers, which arises from using the dominant leg during the game. This is probably the result of more symmetrical actions performed during training and matches, which are due to unexpected situations arising during the game. For both male [21] and female footballers [22], in particular when shooting at goal, which is an important technical skill, there is a clear asymmetry with the right leg being dominant. Barone et al. [23] suspects that sportspeople prefer to use one leg (dominant), because they have better standing balance on non-dominant one.

The regularity of exercising in the context of improving balance has been written about repeatedly. One of the few studies relating to female footballers showed that doing dynamic exercises during warm-up improves both static and dynamic balance [24]. Jakobsen et al. [25] observed postural control improvement after 12 weeks of football training and high-intensity interval running. Paillard et al. [9] paid attention to the fact that the players with more frequent and intensive training sessions showed better postural control, which may be due to greater sensitivity of sensory receptors or better integration of information. Both the general muscular exercises (that mobilizes the whole body) and local muscular exercises (that concentrates on a particular muscle group) can disturb postural control. General muscular exercises (especially running and walking, which are a big part of a football training) contribute to changing the effectiveness of sensory inputs: vestibular—decreases the sensitivity of orthotic organs—and proprioceptive—disturbs the senses of the force and limb position [26]. Söderman et al. [27] Baghbaninaghadehi et al. [28] have drawn attention to the importance of maintaining proper

body stability in team sports. This is an important issue, because footballers must have adequate stability, which is especially important when jumping up to head a ball. Brito et al. [29] found that the postural sway (represented by sway velocity) increased and the ankle proprioception decreased after a 45 min of a match. The authors suggest that prophylactic balance training should be performed following rather than before the practice session. Also, Gioftsidou et al. [30] observed improvement in the balance ability, especially in non-dominant lower limbs when the balance training was performed after the football training.

Mohammadi et al. [31] observed significant changes in both the static and dynamic balance parameters among young sportspeople who undertook a 6-week appropriately selected training schedule. According to the authors, an increase in the sportsperson's muscle mass was responsible for the changes. Results of this work also show better postural stability of young football players in dynamic conditions, which are more valuable than static ones, because they are the closest conditions to these on the football field. Interestingly, Pau et al. [32] observed dynamic balance impairment in young football players due to fatigue after the first half on football match. The impairment was observed in the increase in the time necessary to stabilize after landing from a single-leg jump.

On the basis of tracking professional players' COG trajectory while they were standing on one leg and then monitoring the time taken to stabilize their posture, Pau et al. [32] showed that they achieved significantly lower stabilization times while their center of pressure (COP) displacement surface area was significantly less than for their novice colleagues.

The assessment of the balance and dizziness following sports-related concussions is very important for making decisions about further training or removal from the team [33]. In addition, the study authors [34,35] have shown that the use of exercises to improve balance also has a positive effect on coping with stress and reduces the risk of injury during physical activity, which is extremely important in the effective use of a sportsperson during the game.

The findings of this study must be seen in light of some limitations. The first limitation is the sample size: because of the number of football players in the world it is difficult to deduce a general conclusion based on a small sample. The second limitation concerns the sample profile—using purely student sampling is also extremely limiting on the population. Another limit regarding the student population is the questionnaire and their truthfulness, especially in questions about alcohol, drugs, etc. Gribble et al. [36] observed that males were more adversely affected by fatigue than females. This may suggest that for future research the study group should consist of both sexes. In our research, only dynamic conditions were tested; however, Hrysomallis et al. [37] and Pau et al. [38] paid attention to the fact that assessment of balance in football players should be performed with both dynamic and static tests, because postural control performance is not related in these two cases.

#### **5. Conclusions**

In conclusion, postural stability in young female footballers was found to be better than their inactive peers. The study showed better usefulness of the vestibular system as well as asymmetrical weight-bearing in regularly trainees.

Further research to explore the possibilities of use CDP in assessing the progress of training is needed. The Sensory Organization Test assesses the quality of the human balance system, which could help with evaluation of training effects. Also, the results of the WS may be useful for determination of the dominant leg, which is important when selecting the role the player can play on the field. More studies are needed to specify which exercises contribute to improving the balance and perhaps apply them in rehabilitation techniques.

**Author Contributions:** Conceptualization, G.O.; methodology, G.O.; validation, G.O. and A.C.; formal analysis, A.C.; investigation, A.C..; resources, G.O.; writing—original draft preparation, G.O. and A.C.; writing—review and editing, G.O. and A.C. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Acknowledgments:** We would like to pay our gratitude and our respects to our co-organizer Jozef Bergier, deceased March 2019.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **Evaluation of Commercial Ropes Applied as Artificial Tendons in Robotic Rehabilitation Orthoses**

**Guilherme de Paula Rúbio 1, Fernanda Márcia Rodrigues Martins Ferreira 1, Fabrício Henrique de Lisboa Brandão 2, Victor Flausino Machado 2, Leandro Gonzaga Tonelli 2, Jordana Simões Ribeiro Martins 3, Renan Fernandes Kozan <sup>4</sup> and Claysson Bruno Santos Vimieiro 1,2,3,\***


Received: 28 December 2019; Accepted: 23 January 2020; Published: 31 January 2020

#### **Featured Application: This study can be used as a reference for works that develop active orthoses or prostheses that use artificial tendons to move the fingers, helping in the process of defining these tendons.**

**Abstract:** This study aims to present the design, selection and testing of commercial ropes (artificial tendons) used on robotic orthosis to perform the hand movements for stroke individuals over upper limb rehabilitation. It was determined the load applied in the rope would through direct measurements performed on four individuals after stroke using a bulb dynamometer. A tensile strength test was performed using eight commercial ropes in order to evaluate the maximum breaking force and select the most suitable to be used in this application. Finally, a pilot test was performed with a user of the device to ratify the effectiveness of the rope. The load on the cable was 12.38 kgf (121.4 N) in the stroke-affected hand, which is the maximum tensile force that the rope must to supports. Paragliding rope (DuPontTM Kevlar-<sup>R</sup> ) supporting a load of 250 N at a strain of 37 mm was selected. The clinical test proved the effectiveness of the rope, supporting the requested efforts, without presenting permanent deformation, effectively performing the participant's finger opening.

**Keywords:** orthosis; biomechanics; tensile strength test; robotic therapy

#### **1. Introduction**

Stroke is the world's leading cause of death and disability [1]. with 6.7 million deaths per year, estimated to be the second leading cause of death by the year 2030 [2,3].

Stroke is a clinical syndrome caused by a reduction in blood perfusion in brain structures and is characterized by disturbances in brain function [4]. About 57% of individuals, after having a stroke, have some of limitations in daily activities and more than 50% of these individuals have functional impairment, thus requiring some external help to perform functional tasks [5]. The prevalent motor deficits presented are paralysis (hemiplegia) or weakness (hemiparesis) of body contralateral half to stroke injury [6]. Clinically, muscle weakness, muscle tone abnormality, movement control deficiency, body composition changes of the extremities with loss of muscle and bone mass [7], postural alteration, lack of mobility, abnormal synergistic patterns and loss or reduction of motor coordination can also be observed [8].

An innovative and promising rehabilitation alternative capable of enhancing the motor and functional capacity of individuals after stroke is the robotic therapy [9]. It uses robotic orthoses, which is mechatronic equipment ranging from the creation of artificial limbs, to robots that assist in rehabilitation or hospital or residential care [10].

The great advantage of using these devices is the high degree of repeatability and the performance of intensive activities with less professional supervision, using a simple routine of pre-programmed robot rehabilitation activities [8,11–13]. Numerous systematic reviews have been performed showing the efficiency in using these devices to rehabilitate individuals [12,14–18]. An Improvement in short and long-term proximal end (shoulder and elbow) motor control has been observed in acute and chronic post-stroke patients, using these devices [12,17]. A combination of traditional methods and robotic therapy in some stages of stroke recovery can produce a significant improvement in elbow and shoulder motor recovery [16,18]. An improvement in daily life activity, arm function and muscle strength has also been observed [15], as well as minor effects on motor control and mild effects on muscle strength compared with other short-term interventions using robotic therapy [14]. A combination of these devices with brain–computer interfaces has been also used to improve the rehabilitation ability of individuals after stroke, thus proving the effectiveness of these robots during therapy sessions [19].

Several devices have also been developed to rehabilitate or assist upper limb function, such as MIT-Manus [20], ARMin III [21], MIME [22], BI-MANU-TRACK [23], ARM Guide [24] and NeReBot [25]. In Brazil, the Laboratório de Bioengenharia da Universidade Federal de Minas Gerais (LabBio-UFMG), located in Belo Horizonte/MG, developed a robotic orthosis for upper limb rehabilitation of post stroke individuals [26,27].

Currently is desire the development of soft robots, made by tissue, or elastic polymers [28]. In this way, mechanical tendons or soft actuators act like a human tendon performing the exoskeleton actuation function, while the hand and fingers perform the strutural function [29]. These devices often use springs, cables, elastics, ropes or elastic wires, making the same fuction of artificial tendons, to transmit movement to the paralyzed muscles, through traction and relaxation of them. In the Exo-Glove [29] (Figure 1b) a Bowden cable, frequently used like a bicycle brake cable, was used to perform the artificial tendon function and control the fingers movement, your choice was due to a teflon coating which protects the steel cable. In the Xiloyannis et al. [30,31] work a Bowden cable was used to transmit de motor force to the artificial tendon, but in the first a Kevlar rope was used in the artifical tendon function and in the second another cable steel was used for this. In the Hero system [32] (Figure 1a) a tie up, made by plastic, was used and a simple rope was used in the previously version of the LabBio hand orthesis [33] (Figure 1c).

**Figure 1.** Use of artificial tendon in: (**a**) Hero orthosis, (**b**) hand exoskeleton and (**c**) LabBio orthosis. Adapted from Cherian et al. [29], Yurkewich et al. [32], Rocha [33].

These artificial tendons must be able to withstand heavy loads, have low strain, ease of handling and good fit for the user limb, ensuring safe use of the equipment. To achieve this a correct analysis of each cable must be made, through tests that prove the correct functioning of the tendons. Thus, this paper aims to present the design, selection and testing of commercial ropes that were used to generate the hand movements of a robotic orthosis for upper limb rehabilitation of individuals after stroke.

#### **2. Methodology**

#### *2.1. Device*

The orthosis developed in (LabBio-UFMG) is portable, low cost and low weight composed by a module with glove and artificial fingers and phalanxes, witch are connected to ropes (artificial tendons) that are able to open the user's fingers. The rope is pulled through a power screw system coupled to a motor, which executes and controls the motions. The rope was attached to the transmission system through a support fixed by a bolt (Figure 2).

**Figure 2.** Orthosis with artificial tendons use.

The artificial phalanges limit the users' phalanges movement avoiding anything greater than physiological amplitudes. For fingers' opening movement, the artificial tendon was tensioned and the distal artificial phalanx was pulled, in that way, performing a rotation about the distal articulation axis. This movement was performed until the distal artificial phalanx collided with the middle artificial phalanx. Keeping the traction in the artificial tendon, a rotation about the middle articulation axis was performed until the collision between middle and proximal artificial phalanges. The rotation now occurred about the proximal articulation axis until the transmission system nut actuated the travel limiter sensor present in the system, so the fingers' opening completed movement was performed (Figure 3). The closing movement of the fingers was performed passively, taking advantage of the user's ability to perform this movement. Due to this, the artificial tendon was designed to support the traction during the fingers opening movement.

**Figure 3.** Schematic representation of the (**a**) opening and (**b**) closing fingers movement by the device. Adapted from Rúbio et al. [27].

Before to the operation started, with the orthosis dressed up, the operator commanded the actuator to the fully fingers opening setup with the bracket screw loose.Upon reaching the position, the operator pulled the ropes, with the user's hand fully opening, until they were on full traction, and tightened the fastener screw, thus ensuring the correct performance of the system. All position and speed control was made through an application developed for the Android-<sup>R</sup> system and a microcontrolled electronic circuit by a STM32F103.

#### *2.2. Volunteers*

The volunteers were recruited, for a pilot test with the device, through a public call issued by stroke associations, rehabilitation centers, hospitals and social media in the city of Belo Horizonte, Minas Gerais in December 2016. They were selected according to the inclusion and exclusion criteria and informed about the study objectives. Those who agreed to participate signed the informed consent form. The study project was approved by the Universidade Federal de Minas Gerais Research Ethics Committee (CAAE Registry: 22207213.5.0000.5149).

Inclusion criteria were: age greater or equal than 18 years with left unilateral stroke and chronic impairment diagnosis (minimum six months after stroke) [34]; present hemiparesis with reduced upper limb motor function (elbow and fingers incomplete flexion and extension movements, presenting arm and hand in the 4th Brunnstrom phase) [35]; present at least 45◦ of the shoulder flexion and abduction, complete passive movement of the elbow, hand and finger joints without compromised sensitivity measure by the Fugl–Meyer scale [36]; present muscle tone alteration from mild to moderate, measured through the Modified Ashworth Scale [37]; present absence of severe cognitive deficits, evaluated by the Brazilian version of the Mini-Mental State Examination [38]. The exclusion criteria in the study were: flaccidity in the affected upper limb; severe neurological, orthopedic or rheumatologic impairment before the stroke that may interfere with the performance task; severe cognitive impairment (global aphasia, attention deficit, neglect) that limits understanding of commands or conclusion of experimental tasks; severe pain in the affected upper limb, measured using the Visual Analogue Scale (VAS) (> 8 on a scale from 0 to 10); opened skin injuries where the device would be attached; having used Botox in the last three months for spasticity or other medicines known to increase motor recovery; having participated in the last three months of another research study to improve upper limb function.

Four individuals were selected to participate in the study. Their characteristics are presented in Table 1.


**Table 1.** Volunteers characteristics.

(\*) Spasticity assessed using the Ashworth Scale.

#### *2.3. Artificial Tendons Design and Selection*

For the artificial tendons design and selection used in the hand module, some design steps were required. The first was the determination of the loads to which the ropes were subjected; thereafter a tensile strength test using various commercial ropes to evaluate and determine what would best fit the use; and finally a pilot test with the volunteers to confirm the effectiveness of the rope as an artificial tendon.

#### 2.3.1. Applied Loads Determination

To determine the traction force that the rope would be subjected, direct measurements were performed on the four volunteers with a New Saehan Squeeze Dynamometer-SH5008 bulb-type dynamometer. Normative anthropometric data were not used for this determination because individuals after stroke

present abnormality in tone, which implies hypotonia in the development and subsequent hypertonia in these individuals. This hypertonia can lead to spasticity, which implies speed-dependant increase in muscle tone, thereby increasing myotactic reflex, postural changes and stereotyped movements, which leads to a reduction in the range of joint motion, pain, muscles limb activities limitation and consequently hinders the performance of daily functional activities [39,40]. Because of that, these individuals present a lower grip strength than normative anthropometric data, which present an average of 44.2 kgf for men and 31.6 kgf for women when using the dominant hand [35].

As users of this equipment present difficulties only in the fingers opening movement, the maximum force required to the rope was greater or equal than the maximum grip strength of the hand affected, that is, pulling the rope. The movement of finger closure was given passively, taking advantage of the spasticity presented by the individuals and the ability to close the fingers preserved.

To measure the grip strength, each participant squeezed the dynamometer body as tightly as possible for three consecutive times and an average of the strength measured was calculated. The traction force applied to the selected cable was determined to be the highest of the average forces measured, thus ensuring that the rope safely supports the required loads.

#### 2.3.2. Tensile Strength Test

After defining the applied traction force, the rope used was selected. For this, a set of eight different commercial ropes were subjected to a tensile strength test, until their rupture, to determine the maximum supported force and the strain of each one. The first guarantees the system's operation, and must have a value greater or equal than the volunteers highest average grip force, which was the maximum traction load defined for the project. Already the strain ensures the device a degree of repeatability, since one of the great advantages of robotic therapy is the ability to perform the same movements over and over again on the patient [8,11,12], the selected rope must not have permanent strain after performing one or more actions, to ensure the fingers correct position.

The selected ropes to the tests should have a diameter equal or less than 1.25 mm to easily couple the rope in the nut and artificial phalanges without raising the device components size and weight. Furthermore, they usually used in hight loads applications and could be easily found in the regional market due to the fabrication costs. Their characteristics are presented in the Table 2.


**Table 2.** Ropes characteristic.

The tests were performed at the Structural Analysis Laboratory of the Pontifícia Universidade Católica de Minas Gerais, using the EMIC 23-5D tensile testing machine, which has a maximum load capacity of up 5 KN and clamps for specimen fixation. For the attachment of the ropes, a tie was used, where the cable was tied to the machine's claws ends, using a simple knot. With the cable properly fixed, the upper clamp was adjusted to an initial height of 300 mm, thus allowing a usable wire clearance of 255 mm and a maximum test height of 600 mm.

For control and data acquisition, Tesc Version 3.04 software was used. The test parameters were then defined in the software, which were the "Rectangular Tie Pull" method, the type of material tested and the rope diameter analyzed. The test was initiated by the operator, and the machine gradually increased the applied force automatically until the sample rupture was reached. With the test finished, the software generated the strain and force plots with the value measured. Several tests were performed in order to stabilize the values of force and strain obtained, since the desired curve for each rope should have similar behaviors, without great variability in their force values by strain. However a great variability in the values found were observed, mainly due to the way the ropes were attached to the claw, which often fails to fix them sometimes loosening the knot made which altered the obtained data and thus several tests were needed. The results presented took into account the tests that behaved with greater similarity for each rope.

#### 2.3.3. Pilot Test

With the rope selected, a device functionality test was performed with the four volunteers to approve this rope like artificial tendons. Only the results of one of the volunteers (participant 3) was shown, because this participant presented one of the highest hand spasticity values, so if the tendons could open their fingers efficiently, all of the individuals, with a spasticity lower of equal than his, would be also able to open their fingers.

The opening and closing movement of the fingers (complete extensions and flexions) were performed several times to verify if the artificial tendon supported the loads during the whole movement. Partial openings were also made in order to verify the degree of precision of the system by measuring the opening angle of the hand using a plastic PVC Carci finger goniometer. In addition, it was verified that the cable did not suffer any permanent strain during its operation.

#### **3. Results**

#### *3.1. Applied Loads Determination*

With the data of volunteers grip strength (Table 3), the highest average grip strength observed was equal to 17.83 kgf (174,91 N) in the hand afected by the stroke. Therefore this is the maximum traction force that the rope used should support.


**Table 3.** Unaffected and affected limb grip strength of volunteers.

#### *3.2. Tensile Strength Test*

The TufLine XP fishing line behaved heterogeneously, as shown in the three force x strain curves below (Figure 4). It presented the maximum strength range between 45 and 140 N and a strain between 14 and 39 mm approximately.

**Figure 4.** Force x strain curves TufLine XP fishing line.

The guitar strings tested, showed a more consistent behavior, and were analyzed as shown in Figure 5. Only one curve force x strain for each string was shown, due to close behavior of the curves obtained by the tests. Among the guitar strings submitted to the tensile strength test, the one that presented the best result was the D string (0.64 mm in diameter), it presents approximately a 37 mm of strain, when it is pulled at 200 N force. This value is higher than the global average grip force in the volunteers (12.38 kgf or 121.45 N) and the maximum traction force determined in the project (174.91 N). The ropes did not perform as expected, there was no proportional increase in the maximum tensile strength relative to the increase in the diameter of the rope. This is because, in order to change the musical notes (frequencies), in addition to increasing the diameter, changes in the string structures are necessary.

**Figure 5.** Force x strain curves of: **Guitar String E, Guitar String B, Guitar String G, Guitar String D, Guitar String A, Guitar String E (1.14 mm)**.

The last rope analyzed was the SOL Paragliders paragliding rope. The force x strain curves shown (Figure 6) to demonstrate the more stable behavior of the paragliding rope than the fishing line. It withstood high tensile loads between 225 N and 280 N and maximum strain between 33 mm and 35 mm. In addition, the generated curves present greater congruence of values for the elastic, plastic and rupture limit regions compared to the other lines submitted to the test.

**Figure 6.** Force x strain curves of paragliding rope of SOL Paragliders.

#### *3.3. Pilot Test*

With the artificial tendon selected, it was then possible to use it with a post stroke individual to validate its application. As shown in Figure 7, the participant's finger opening, using the orthosis, was satisfactory, the full opening was performed, which was not possible naturally, through voluntary control. As shown in the Figure 7a, the spasticity prevented the volunteer's hand opening movement, which at the same time performed wrist pronation due to stroke sequelae. With the orthosis (Figure 7b), pronation was no longer performed and the full opening movement of the hand was performed, proving the functionality of the device for this individual.

**(a) (b) Figure 7.** Participant's finger openings of (**a**) naturally form and (**b**) using the orthosis.

Making repeated partial openings of 80◦, 60◦ and 40◦ (Figure 8), it was observed that the fingers presented a consistency and precision in the opening angle, showing that the use of this type of rope can satisfactorily perform as an artificial tendon in robotic orthoses for individuals after stroke. It is important to emphasize that the repeated use of the equipment did not generate permanent strains in the rope, being able to execute without variations the same amount of opening requested.

**(a) (b) (c) Figure 8.** Fingers aperture at (**a**) 80◦ , (**b**) 60◦ and (**c**) 40◦.

#### **4. Discussion**

Results of the volunteers grip force measures confirm that they have a strength lower than that of typical individuals. Normative data indicate that healthy individuals had an average dominant hand grip force of 28.4 ± 9.7 kgf for women and 40.3 ± 14.3 kgf for men [41]. The change in post-stroke individuals grip force is related to the motor deficits found after the brain injury, which result from the injury to the upper motor neurons that control the distal and proximal muscles, generating a decrease in the activation of some muscle groups [42]. In addition, there is the presence of spasticity, could cause postural changes in the upper limb, which, if not corrected it generate deformities, affecting even the joints and muscles viscoelastic properties and the tendons integrity. Therefore, based on the studies of Wu et al. [43,44], we can use a high resolution ultrasound as an alternative to check these phenomena and select the most suitable cable to be used in robotic orthoses.

It can be noted in the fishing line tensile strength test (Figure 4), that there is a variability in its behavior, presented maximum traction strength below than the desired. In addition to inconsistent behavior, it is difficult to handle and could difficulty in properly adjusting tendons before orthosis operation.

The D'Addarío (85/15 bronze) guitar strings (Figure 5), despite their smallest strain, suffered some damage with the application of low force, once their constitution was related to the sound and not to the material resistance. Its structure broke internally or suffered cracks with low loads, which can be seen with a simple visual inspection.

The paragliding rope showed the highest tensile strength among the tested ropes (Figure 6), breaking through the external structure destruction, with a smoothly break.

Thus, given the analysis of data obtained in the tests performed with different ropes, the conclusion was that the most suitable for use in robotic orthosis is the paragliding rope which meets the pre-established criteria of: strength, diameter and cost. (DuPontTM Kevlar-<sup>R</sup> ). The rope presented an easy handling and greater security due to values of resistance to traction superior to the other ropes, and due to its form of rupture, since the others present a whip when broken.

The pilot test confirmed the consistency and precision in the opening angle using this rope, showing the satisfactory use as an artificial tendon in robotic orthoses. This system makes it possible to efficiently open and close the post-stroke individuals fingers, since a significant proportion of individuals with physical sequel resulting from stroke, remains with problematic or unsatisfactory manual function return [45]. With the use of artificial tendons in the orthosis, there is an increase in functional skills, facilitating the daily activities performance, such as reaching, picking up and holding objects, using tools such as a cell phone, eating, dressing and performing personal care in a general way. This significantly impacts the patient's life, improving his independence, self-esteem and life quality.

#### **5. Conclusions**

The use of artificial tendons applied to upper limb orthoses is one of the ways to effectively perform fingers opening of the individual. As shown in this study, commercial ropes can be used for this purpose as long as they support the requested loads without presenting a permanent strain. We analyzed eight ropes to act as artificial tendon and withstand a tensile load of 121.4 N. Of these, the paragliding rope manufactured by SOL Paragliders showed the best performance for this purpose, withstanding a load of 250 N at approximately 30 to 35 mm of strain. When used in the device, in an individual after stroke, it was effective for application, being able to perform the task without breaking, and maintaining the degree of repeatability, i.e., did not suffer permanent strain, which would affect the equipment degree of repetition, validating its use as an artificial tendon.

**Author Contributions:** Conceptualization: G.P.R., F.M.R.M.F. and C.B.S.V.; Data curation: F.H.d.L.B., V.F.M., L.G.T.; Writing and editing: G.d.P.R., F.M.R.M.F., J.S.R.M. and C.B.S.V.; Supervision: R.F.K. and C.B.S.V.; Project administration: C.B.S.V. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by Financiadora de Estudos e Projetos (FINEP: 01.12.0476.00), Fundação de Amparo à Pesquisa do Estado de Minas Gerais (FAPEMIG) and Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (Capes): finance code 001.

**Acknowledgments:** The authors would like to thank the Universidade Federal de Minas Gerais, Pontifícia Universidade Católica de Minas Gerais and the Graduate Program in Mechanical Engineering for the support available to carry out this project.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


c 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **Texture Analysis is a Useful Tool to Assess the Complexity Profile of Microcirculatory Blood Flow**

#### **Henrique Silva 1,2,\*, Hugo A. Ferreira 3, Clemente Rocha <sup>1</sup> and Luís Monteiro Rodrigues 1,2**


Received: 31 December 2019; Accepted: 28 January 2020; Published: 30 January 2020

**Abstract:** The quantitative assessment of cardiovascular functions is particularly complicated, especially during any physiological challenge (e.g., exercise), with physiological signals showing intricate oscillatory properties. Signal complexity is one of such properties, and reflects the adaptability of the physiological systems that generated them. However, it is still underexplored in vascular physiology. In the present study, we calculate the complexity of photoplethysmography (PPG) signals and their frequency components obtained with the wavelet transform (WT), with two analytical tools—(i) texture analysis (TA) of WT scalograms, and (ii) multiscale entropy (MSE) analysis. PPG signals were collected from twelve healthy young subjects (26.0 ± 5.0 y.o.) during a unilateral leg lowering maneuver to evoke the venoarteriolar reflex (VAR) while lying supine, with the contralateral leg remaining stationary. Results showed that TA was able to detect a decrease in complexity, viewed as an increase in texture entropy (TE), of the PPG scalograms during VAR, similarly to MSE, suggesting that a decrease in the competence of vascular regulation mechanisms might be present during VAR. Nonetheless, TA showed lower sensitivity than MSE for low frequency spectral regions. TA seems to be a promising and straightforward analytical tool for the assessment of the complexity of PPG perfusion signals.

**Keywords:** signal complexity; texture analysis; multiscale entropy analysis; wavelet transform; photoplethysmography

#### **1. Introduction**

The regulation of the cardiovascular system results from the adjustment of several biophysical phenomena, both electrical and mechanical. Their coordination is highly complex, depending on the continuous feedback and cross-talk between effector organs and controlling systems [1]. The heart pump and respiration are the most notorious "central" processes governing the performance of the cardiovascular system, while on a more "peripheral" level, the phenomena that control vascular tone, such as the myogenic activity of the vessel wall, the sympathetic activity and the endothelial release of vasoactive substances, play crucial roles [2–4]. Microcirculation signals are composed of both these central and peripheral components and therefore provide an "integrated" view of cardiovascular function [4,5]. The contributions from these multiple regulation systems and their intricate interplay explain the oscillatory properties of microcirculation signals. From an analytical standpoint, these oscillatory properties are increasingly considered as a means to extract more sensible information

not apparent from the general analysis of the raw signals, in particular their spectral origin and the significance of their fractal and chaotic profiles [5–7]. In recent years, much attention has been given to the study of the complexity of physiological signals, a property that reflects the adaptability of the systems that generated them. Intuitively, the term complexity is often associated with "structural richness" [8], which can be applied to the study of biological signals and images. Complexity is typically assessed as "entropy," a general measure of "disorganization" in physical, chemical and biological phenomena [9]. In biomedical research, entropy is used as a quantitative parameter, and is typically assessed in continuous, but temporally unpredictable, physiological signals, such as electrocardiography [10], electroencephalography [11], blood pressure [12], laser Doppler flowmetry (LDF) [13], and photoplethysmography (PPG) [14]. Physiological signals may be regular (e.g., more periodic) or irregular. The higher entropy measured in irregular signals reflects the adaptation capability to changing internal and external conditions of the biological system that generated them [15]. Overall, the entropy of physiological signals tends to decrease whenever the systems that generated them lose adaptability, which can result in a compromise of their function or even in disease. Therefore, a suitable metric of complexity should assign higher values to the output signal of a "healthier" system with a rich and meaningful structure and lower values to either random dynamics or predictable systems, which are often associated with disease [8,16,17]. This has been observed in the aging process [18,19] and in cardiovascular, metabolic and neurological diseases [18], among several others. In addition to pathological states, some experimental procedures are sufficiently "challenging" to temporarily provoke tissue dysfunction, especially those that reduce perfusion [6]. For continuous signals, the multiscale entropy analysis (MSE) is considered a robust analytical tool, having shown superiority over other measures for its ability to assess entropy over different time scales [19]. Entropy can also be assessed in texturally rich biological images, reflecting their textural "disorganization." Texturally rich images can be assessed with texture analysis (TA), an analytical tool allowing the detection of various features on a gray-level image to discriminate textural differences, such as texture entropy (TE), contrast, correlation, homogeneity, and energy. Entropy calculated from TA of images may reflect different information than that conveyed by the entropy assessment of continuous numerical series. When applied to a grayscale image, high entropy means that the pixels can adopt a high number of gray levels, expressing richer information [20]. TA has been used in medical imaging with the purpose of improving diagnostic capacity, given that human visual inspection constitutes a process that is expensive, time-consuming and prone to interpretation errors [21]. Thus far, TA in medical imaging has been more frequently applied to mammography [22], ultrasonography, computerized tomography (CT) [23], and magnetic resonance imaging (MRI) [24] techniques [25], as well as to dermatoscopy [26] and microscopy images [27]. In particular, TE has been used to explore immune competence of lymphoid tissue in microscopic images [28], to assess the onset organ (liver) disease and tumor (colorectal cancer) evolution in CT scans [29–31], and to assess bone regeneration in fractures [32] and osteoarthritis [33], among others. In medical imaging, TE does not necessarily decrease when images display abnormal features. In fact, since abnormal or dysmorphic features often increase the coarseness of smooth or uniform images from healthy subjects, increases in TE are often seen in pathological images, such as in the case of aged tissues and organs [34] as well as the case of neoplastic formations [30,31]. In these situations, a richer texture and therefore, higher entropy, is associated with dysfunction. Although currently underexplored in vascular physiology, the complexity assessment of microcirculation signals, both raw and decomposed, may deepen our knowledge of the mechanisms underlying perfusion regulation. Microcirculation signals are easily decomposed into their frequency components by the wavelet transform (WT) [4]. One of the WT main outputs is a scalogram, i.e., a multi-patterned image representation of the time evolution of all components of the decomposed signal. Given their textural richness, WT scalograms seem suitable candidates for TE analysis, however, to our knowledge this analysis has not been previously attempted. Since WT scalograms are not medical images, the interpretation of TE in this context must be necessarily different from the one performed in medical diagnosis. Given the fact that a WT scalogram is a 2D projection of

a 3D frequency spectrum, (i.e., the 3D spectrum seen from above), it is only logical to assume that a change in the entropy of the numerical series that constitute the 3D spectrum would also be observed in corresponding WT scalograms. Therefore, the assumption that similar trends in complexity would be observed with MSE and TA seems legitimate. Our objective was to compare the complexity of microcirculation PPG signals calculated with TE against MSE during a typical venoarteriolar reflex (VAR).

#### **2. Materials and Methods**

#### *2.1. Experimental*

Twelve healthy young adult subjects (26.0 ± 5.0 y.o., seven females, five males) participated in this study after giving informed written consent. All subjects were healthy, with no cardiovascular diseases, were non-smokers, and abstained from consuming alcohol and caffeine-containing beverages 24 hours prior to the experimental procedure. The protocol was approved by the School of Health Sciences and Technologies' Ethical Commission. All subjects gave written informed consent, and the study was conducted in accordance with the Declaration of Helsinki and subsequent amendments [35]. After 20 minute acclimatization to room conditions (temperature: 22 ± 1 ◦C, humidity: 40–60%), a postural challenge to elicit the VAR was applied, as previously described [4]—10 minutes lying supine with both legs extended (baseline, Phase I); 10 minutes with one foot lowered 50 cm from heart level (challenge, Phase II); and 10 minutes resuming the initial posture (recovery, Phase III). The contralateral foot remained stationary and served as control. Blood flow signals, expressed in arbitrary units (AU), were acquired from the first toe of both feet at a 100 Hz sampling rate with a reflection photoplethysmography Blood Pulse Volume sensor (Biosignals Plux, Lisboa, Portugal) connected to a BITalino Plugged microprocessor board (Biosignals Plux).

#### *2.2. Numerical*

The raw PPG signals were imported to Matlab (Mathworks R2012, Natick, MA, USA) and smoothed with a moving average filter.

The WT (http://noc.ac.uk/using-science/crosswavelet-wavelet-coherence) was then applied to the smoothed signals, which allowed the decomposition into their main frequency components. Wavelets are a family of functions constructed from translations and dilations of a single function called the "mother wavelet" ψ(t), collectively defined by:

$$
\psi\_{\mathbf{a},\mathbf{b}}(\mathbf{t}) = \frac{1}{\sqrt{|\mathbf{a}|}} \psi\left(\frac{\mathbf{t} - \mathbf{b}}{\mathbf{a}}\right), \mathbf{a}, \mathbf{b} \in \mathbb{R}, \text{ a } \neq 0 \tag{1}
$$

The parameter *a* is the scaling parameter or scale, measuring the degree of compression, while parameter *b* is the translation parameter, determining the time location of the wavelet. If |a| < 1, then the wavelet in the above equation is the compressed version of the mother wavelet and corresponds mainly to higher frequencies. If |a| > 1, then ψa,b(t) has a larger time-width than ψ(t) and corresponds to lower frequencies. Thus, wavelets have time-widths adapted to their frequencies.

In this case, the central frequency of the wavelet was 6 Hz, with a discretization factor of 10 scales per octave. The detected components of the PPG signals occurred at the following frequency ranges: cardiac [1.6–0.7 Hz], respiratory [0.7–0.26 Hz], myogenic [0.26–0.1 Hz], neurogenic/sympathetic [0.1–0.045 Hz], endothelial NO-dependent (NOd) [0.045–0.015 Hz] and NO-independent (NOi) [0.015–0.007 Hz], which were in line with the ranges previously described for LDF [36]. From the WT, two main outputs were generated for each subject: (1) a scalogram and (2) a 3D periodogram. The WT scalogram (Figure 1) shows the time (here converted to sample number) evolution of the amplitude with a factor of 2<sup>n</sup> (here in color scale) for each signal scale. The approximate frequency of a component is obtained by dividing the central wavelet frequency by the scale of the component. From a visual inspection, the WT scalogram also presents an oscillatory pattern in pixel distribution on different scales, making it a

suitable candidate for TA. The scalogram was then used as a grayscale image, where several regions of interest (ROI) were marked, each per component per phase of the protocol. Each ROI was then converted to a gray-level co-occurrence matrix (GLCM), a matrix where the number of rows and columns is equal to the number of gray levels (G) in the image, from which the TE was calculated as follows [37]:

$$\text{Entropy} = -\sum\_{\mathbf{i}=0}^{G-1} \sum\_{\mathbf{j}=0}^{G-1} \mathbf{P}(\mathbf{i}, \mathbf{j}) . \log[\mathbf{P}\left(\mathbf{i}, \mathbf{j}\right)] \tag{2}$$

where *P*(*i*, *j*) denotes the probability of occurrence of a given element in the matrix.

**Figure 1.** WT scalogram of the PPG signal from the control (top) and test (bottom) feet from a representative subject (20 y.o.).

To correctly select each ROI, a control sine wave signal was generated with multiple components occurring at known frequencies (f1 = 3.2 Hz; f2 = 1.6 Hz; f3 = 0.7 Hz, f4 = 0.26 Hz; f5 = 0.1 Hz; f6 = 0.045 Hz; f7 = 0.015 Hz; f8 = 0.007 Hz), where each frequency is close to the ones that define the borders of the PPG components' detected frequency ranges. The sine wave signal was then deconstructed with the WT, and a reference scalogram was generated (Figure 2).

From the 3D periodogram (Figure 3), a 2D time evolution of the amplitude of each component was constructed by averaging the amplitude at each time point for each spectral interval (Figure 4). Signals were acquired at a 100 Hz sampling rate, which translates to a total number of 180,000 samples for the entire 30 minute protocol. The wavelet period is a natural logarithmic representation of the wavelet scale, with the frequency in Hz exponentially related to this period. The time evolution was

analyzed for both the raw PPG signal and its components with MSE, an algorithm that quantifies the randomness/unpredictability of a signal over different time scales as follows [19]:

**Figure 2.** WT scalogram of the reference sine wave signal. The region corresponding to each component frequency interval is shown (1—first harmonic of the cardiac component; 2—cardiac; 3—respiratory; 4—myogenic; 5—sympathetic; 6—endothelial NO-dependent; 7—endothelial NO-independent).

**Figure 3.** PPG 3D frequency spectrum (period vs. sample vs. amplitude) for the test foot of a representative subject (20 y.o.). Signals were acquired at a 100 Hz sampling rate, which translates to a total of 180,000 signal samples for a 30 minute acquisition period. Period is a natural logarithmic representation of the wavelet scale.

Given a time series *xi* = 1,..., *N*, a consecutive coarse-grained time series *y*(τ) is constructed:

$$y\_j^{(\tau)} = \frac{1}{\tau} \sum\_{i=(j-1)\tau+1}^{j\tau} x\_i \tag{3}$$

where τ represents the scale factor and 1 ≤ j ≤ N /τ. The sample entropy (*SampEn*) of each coarse-grained time series is then computed. *SampEn(m, r, N)* is the negative natural logarithm of the conditional probability that a dataset of length *N*, having repeated itself with a tolerance of *r* points, will also repeat itself for *m* + 1 points, without allowing self matches:

$$\text{SampEn}(\mathbf{m}, \mathbf{r}, \mathbf{N}) = -\ln \frac{\mathbf{A}^{\mathbf{m}}(\mathbf{r})}{\mathbf{B}^{\mathbf{m}}(\mathbf{r})} \tag{4}$$

where *Am(r)* is the probability that two sequences will match for *m* + 1 points and *Bm(r)* is the probability that two sequences will match for *m* points. The more regular and predictable a time series is, the lower the value of *SampEn*. The more random a time series is, the higher the value of *SampEn*. Plotting the *SampEn* over the scale factor yields the MSE curve, which gives insight into the integrated complexity of the system over the time scales of interest, which can be of interest when comparing groups where differences in specific time scales are probable. The randomness/unpredictability of the signal can finally be straightforwardly summarized as the complexity index (CI), which corresponds to the area under the MSE curve [38]. The CI and TE were statistically compared between each phase of the protocol with the Wilcoxon signed-rank test, and were compared between feet for each phase with the Mann-Whitney independent sample test, adopting a 95% confidence interval.

**Figure 4.** Time evolution of the PPG signal components' amplitude ratio for the control (blue) and test (red) feet throughout the protocol (baseline: signal samples 0–60,000; challenge: signal samples 60,000–120,000; recovery: signal samples 120,000–180,000) for a representative subject (20 y.o.).

#### **3. Results**

As previously published, during the postural challenge, perfusion decreased significantly on both the test and control feet [4]. The mean and standard deviation (SD) of the entropy parameter (CI) calculated with MSE is presented in Table 1. For the test foot, the CI of the raw PPG signal and all its components decreased significantly during the challenge (Phase II, raw signal: *p* = 0.005; cardiac: *p* = 0.012; respiratory: *p* = 0.002; myogenic: *p* = 0.003; sympathetic: *p* = 0.002; NOd: *p* = 0.002; NOi: *p* = 0.002). For the control foot, the CI of the raw PPG signal showed an increase without statistical significance. All the components' CI decreased, with the exception of the cardiac, although not so pronounced as in the test foot (respiratory: *p* = 0.002; myogenic: *p* = 0.003; sympathetic: *p* = 0.002; NOd: *p* = 0.002; NOi: *p* = 0.002). During recovery (Phase III), significant differences were still detected on both feet regarding baseline for the respiratory (test: *p* = 0.002; control: *p* = 0.002), myogenic (test: *p* = 0.008; control: *p* = 0.008), sympathetic (test: *p* = 0.002; control: *p* = 0.002), NOd (test: *p* = 0.002; control: *p* = 0.002) and NOi (test: *p* = 0.002; control: *p* = 0.003). The mean and SD of the entropy parameter (TE) calculated for the sine wave and PPG signals with TA is shown in Table 2. For the sine wave

signal scalogram, TE decreased from the cardiac component towards the low frequency components, where it increased again for the NOi. For the test foot PPG scalogram, TE decreased significantly in the cardiac (*p* = 0.004), respiratory (*p* = 0.013), and myogenic (*p* = 0.008) components, increased significantly in the NOd (*p* = 0.012) and NOi (*p* = 0.010), and showed no change in the sympathetic component. During recovery, all components' TE returned except for the NOd, which showed a significant difference regarding baseline (*p* = 0.007). For the control foot PPG scalogram, changes in TE were not so pronounced as in the test foot, with only the cardiac component showing a significant difference regarding baseline (*p* = 0.001). During recovery, only the cardiac component showed a significant difference (*p* = 0.022) regarding baseline. No significant differences in entropy were found between feet at baseline (Tables 1 and 2) either with CI or TE. During the challenge phase, significant differences were found for the raw signal (*p* < 0.001) and all components (p=0.001 for the respiratory; *p* < 0.001 for the remaining) with CI. Similarly, TE showed significant differences in all components (cardiac: *p* = 0.001; respiratory: *p* = 0.045; myogenic: *p* = 0.028; NOd: *p* = 0.001; NOi: *p* = 0.001), except the sympathetic. During recovery, the CI of the sympathetic (*p* = 0.012), NOd (*p* = 0.002) and NOi (*p* < 0.001) were significantly different, while with TE no statistical differences were found.

**Table 1.** Mean and standard deviation (SD) of the MSE complexity index for each foot on each phase of the protocol (card—cardiac, resp—respiratory, myo—myogenic, sym—sympathetic, NOd—endothelial NO-dependent, NOi—endothelial NO-independent). Statistical comparison to Phase I is shown (\*—*p* < 0.05).



**Table 2.** Mean and standard deviation (SD) of the texture entropy for each foot on each phase of the protocol (card—cardiac, resp—respiratory, myo—myogenic, sym—sympathetic, NOd—endothelial NO-dependent, NOi—endothelial NO-independent). Statistical comparison to Phase I and between the test and control limbs are shown (\*—*p* < 0.05).

#### **4. Discussion**

In this study, the acquired PPG signals and an artificial sine wave signal were decomposed with the WT into their respective components so that TE and CI could be calculated. The relative contributions of each of the components to the overall signal, expressed as amplitude ratios, are typical WT outputs but were not considered for this study as these results were previously published [4]. During leg lowering, perfusion decrease in the test foot is explained by a constriction of arterioles secondary to the increased venous distension when the foot is in a pendent posture, which constitutes the VAR [39]. The fast perfusion decreases in the control foot (not so pronounced as in the test foot) is thought to be due to a centrally mediated neurogenic reflex initiated to maintain the vascular homeostasis in the lower limb [4]. The sine wave signal scalogram showed TE values generally invariant regarding the component, which reflects the regularity of these signals. The highest TE values were recorded for both the highest (cardiac) and lowest (NOi) frequency components, meaning a richer texture was found in the corresponding ROIs, likely explained by the particular overlap of the spectral waves that define them. Contrarily, the PPG signal scalogram showed that the higher frequency (HF) components (cardiac, respiratory, myogenic) displayed higher TE values in either foot, meaning the corresponding ROIs are more texturally rich, i.e., more complex in comparison to the lower frequency components. This suggests that TE assessment of PPG scalograms is sensitive to the underlying phenomena that explain the physiological complexity of these signals. In our data, both TE and MSE analyses showed an overall decrease in the PPG signal and in the respective components' entropies during leg lowering in both the test and control feet. This result seems to indicate a bilateral decrease in the vascular adaptation capacity to the lower perfusion levels observed during leg lowering. This apparent decrease in adaptation capacity is further reinforced by the observation that several PPG components' entropies show significant differences between the recovery and baseline phases, suggesting that full recovery was not attained at the end of the protocol. In our view, these results indicate that that VAR creates a state of reduced perfusion that, while not compromising tissue viability, destabilizes the evoked regulatory mechanisms. When considering the test foot, MSE detected a significant decrease in the entropy of all components during VAR. TE, however, detected an entropy decrease in the cardiac,

respiratory, and myogenic components, an increase in both endothelial components and no change in the sympathetic. Regarding the control foot, an overall entropy decrease was observed with both tools. MSE recorded a decrease in the complexity of all components, although not so pronounced as in the test foot, which again is likely attributable to the magnitude of the perfusion decrease. Again, TE showed more subtle changes, only significant for the respiratory and NOi components. Overall, MSE and TE showed similar responses in the high frequency components (cardiac, respiratory, myogenic), and dissimilar responses in the low frequency components (sympathetic, NOd, NOi), suggesting that TE loses analytical sensitivity in the latter regions. This loss of sensitivity may be attributed to an incomplete visual resolution of the scalogram bands of the low frequency components, leading to a less precise identification of the respective ROIs. Nevertheless, TE was able to differentiate the entropies of most components during VAR between the test and control limbs. Overall, although both TE and MSE were able to detect differences in entropy, MSE was consistently more sensitive. One should keep in mind that MSE and TE ensure different scale evaluations—frequency for TE and time for MSE, which are two aspects of the same reality (i.e., perfusion signal). Thus, both analytical strategies provide different views into the same physiological event, which can also help to explain the differences in sensitivity found. MSE detected changes in entropy between limbs during both VAR and recovery phases, while TE was only able to detect the more pronounced changes that occurred during VAR. Furthermore, MSE was sensitive to changes in entropy for a greater number of components in comparison to TE. Although the performance of the reference MSE method was superior under the conditions studied, these results suggest that TE is an interesting and suitable tool for assessing the complexity of PPG microcirculation signals considered as WT scalograms, in particular for the analysis of high frequency components.

#### **5. Conclusions**

In this paper we present for the first time the use of texture entropy (TE) for the quantification of complexity in microcirculatory perfusion signals, using MSE as a reference analytical tool. Our results show that TE was able to detect a decrease in complexity of the PPG scalograms during VAR, similarly to MSE. This complexity decrease suggests that a decrease in the competence of vascular regulation mechanisms might be present during VAR and should be further investigated. Results from both tools were aligned for the high frequency components, but not for the low frequency components, suggesting a decrease in sensitivity of TE in the latter spectral regions. Recognizing the value of increasing the sample size (and heterogeneity) in future studies, our results have shown Texture Analysis (TA) to be a promising method to assess the complexity of PPG perfusion signals.

**Author Contributions:** Conceptualization, H.S.; Methodology, H.S., C.R. and H.A.F.; Software, H.S. and H.A.F.; Validation, H.A.F., and L.M.R.; Formal Analysis, H.S.; Investigation, H.S. and C.R.; Resources, H.A.F. and L.M.R.; Data Curation, H.S. and H.A.F.; Writing—Original Draft Preparation, H.S.; Writing—Review & Editing, H.A.F.; Visualization, H.A.F. and L.M.R.; Supervision, L.M.R.; Project Administration, L.M.R.; Funding Acquisition, L.M.R. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was supported by national funds from FCT—Fundação para a Ciência e a Tecnologia, I.P, within the project UID/DTP/04567/2019.

**Acknowledgments:** The authors would like to express their thanks to all the volunteers for their participation in this study.

**Conflicts of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

**Ethics Statement:** This study was carried out in accordance with the recommendations of Ethical Principles for Medical Research Involving Human Subjects, Declaration of Helsinki. The protocol was approved by the School of Health Sciences and Technologies' Ethical Commission. All subjects gave written informed consent in accordance with the Declaration of Helsinki.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **The 'DEEP' Landing Error Scoring System**

#### **Kim Hébert-Losier 1,\*, Ivana Hanzlíková 1, Chen Zheng 2, Lee Streeter <sup>3</sup> and Michael Mayo <sup>2</sup>**


Received: 20 December 2019; Accepted: 24 January 2020; Published: 29 January 2020

**Featured Application: The Landing Error Scoring System, an injury-risk screening tool used in sports to detect high risk of anterior cruciate ligament injury, can be automated using deep-learning-based computer vision on 2D videos combined with machine learning methods. The successful application of this method paves the way for the automatic detection of individuals at high risk of injury using smartphone-based applications and opens doors to addressing other related injury prevention problems.**

**Abstract:** The Landing Error Scoring System (LESS) is an injury-risk screening tool used in sports; but scoring is time consuming, clinician-dependent, and generally inaccessible outside of elite sports. Our aim is to evidence that LESS scores can be automated using deep-learning-based computer vision combined with machine learning and compare the accuracy of LESS predictions using different video cropping and machine learning methods. Two-dimensional videos from 320 double-leg drop-jump landings with known LESS scores were analysed in OpenPose. Videos were cropped to key frames manually (clinician) and automatically (computer vision), and 42 kinematic features were extracted. A series of 10 × 10-fold cross-validation experiments were applied on full and balanced datasets to predict LESS scores. Random forest for regression outperformed linear and dummy regression models, yielding the lowest mean absolute error (1.23) and highest correlation (*r* = 0.63) between manual and automated scores. Sensitivity (0.82) and specificity (0.77) were reasonable for risk categorization (high-risk LESS ≥ 5 errors). Experiments using either a balanced (versus unbalanced) dataset or manual (versus automated) cropping method did not improve predictions. Further research on the automation would enhance the strength of the agreement between clinical and automated scores beyond its current levels, enabling quasi real-time scoring.

**Keywords:** anterior cruciate ligament; automation; drop jump; injury risk; deep learning; machine learning; movement screen; OpenPose

#### **1. Introduction**

Lower-extremity injuries due to physical activities have devastating short-term and long-term consequences to the health and wellbeing of individuals [1,2] and burden societies worldwide [3,4]. Non-contact injuries account for approximately 20% of injuries in game situations and 37% of injuries in training situations [5]. Non-contact injuries in sport and recreation are the ones of most practical interest to coaches and clinicians as preventable through neuromuscular training programs [6].

The mechanism of non-contact lower-extremity injuries and their underlying risk factors have been linked with 'risky' movement patterns [7,8], such as knee valgus and stiff landings. 3D motion analysis systems, which provide gold-standard measures for the objective quantification of human motion noninvasively, can readily identify altered movement patterns and biomechanical control. However, conventional 3D motion analysis using infrared systems requires a considerable financial outlay and an expert-user, in addition to time and space to perform the analysis. These constraints limit its practical application and use for large-scale screening of injury risk factors in physically active individuals.

As a countermeasure and to reduce technological requirements, various clinician-led movement screens have been developed [9]. Even though these clinician-led screens reduce the financial costs and space requirements compared to 3D motion analysis, they nonetheless require expert clinicians and dedicated time for testing and scoring, limiting their widespread use. For instance, the Functional Movement ScreenTM takes 12 to 15 min and the Tuck jump assessment takes 12 min to administer and score for one individual [9].

The Landing Error Scoring System (LESS) is one movement screen with demonstrated reliability [10,11] and validity [11,12]. Clinicians evaluate 2D video recordings from three double-leg drop-jump landing tasks per individual to detect 'movement errors' linked to non-contact anterior cruciate ligament (ACL) and other lower-extremity injury mechanisms [10]. The LESS consists of 17 items (Table 1), with the total number or possible errors ranging from 0 (best) to 17 (worst). Greater scores hence indicate more movement errors, poorer landing biomechanics, and greater relative risk of sustaining non-contact lower-extremity injuries. In a prospective study, Padua et al. [12] determined that scoring 5 or more errors on the LESS was associated with a 10.7 times greater relative risk of sustaining a non-contact ACL injury in youth soccer players (sensitivity 0.86, specificity 0.64). The total testing time (including set up) takes ~5 min with 3 to 4 min for a trained rater to score the three drop-jump landing trials of one individual once downloaded to a computer [10].

A few of the drawbacks of the LESS is the subjective nature of the assessment, requirement for an expert-rater, and need to view videos at a later stage [13,14]. In recent years, researchers have striven to automate the LESS to streamline the process using depth sensor cameras [13,15]. Dar, Yehiel, and Cale' Benzoor [13] introduced the PhysiMax system (PhysiMax Technologies Ltd., Tel Aviv, Israel) to automate LESS scoring using a personal computer, 3D Microsoft Kinect, and motion analysis software that requires limited clinical input. Their results indicated high consensus between clinician and PhysiMax LESS scores (intra-class correlation, ICC = 0.80, mean absolute difference 1.13 errors), although the clinician manually inputted the overall impression item (no. 17, Table 1). Despite the automated quantification of the LESS using markerless motion capture using depth cameras provides time-cost saving benefits, there are still additional hardware-software expenditures to consider.

Deep-learning-based computer vision technologies enable the automatic identification and quantification of human motion without the need for depth sensor cameras. Numerous such systems are currently being developed. For example, OpenPose [16] is a system enabling real-time multi-person pose estimation in video streams captured by a camera. The system tracks both body pose as well as keypoints associated with joints and anatomical features. The same technology is also being deployed for solving other related problems, such as tracking lab animal motion in laboratory settings [17,18]. In this work, we aim to apply deep-learning techniques to LESS score estimation. Applying these approaches to 2D video recordings would improve the accessibility to end-users and pave the way to smartphone-based applications for injury risk screening. Our aim is to evidence that LESS scores can be automated from 2D videos using deep-learning-based computer vision with machine learning and compare the accuracy of LESS predictions using different video cropping and machine learning methods. Our work substantiates that: LESS automation is possible without the need for 3D motion analysis or depth sensor cameras, random forest leads to more accurate predictions than linear or dummy (ZeroR) regression models, and that cropping method (manual versus automated) does not affect predictions.


**Table 1.** Landing Error Scoring System operational definitions of errors. (Adapted from Padua et al. [10].)

*Abbreviations*: IC, initial contact; KFmax, maximal knee flexion.

#### **2. Materials and Methods**

#### *2.1. Participants*

A sample of 144 individuals (45 males and 99 females) volunteered to participate in this study. Age, height, mass, and body mass index (mean ± standard deviation) for males were 21.0 ± 5.9 years (range 17 to 42 years), 179.1 ± 7.2 cm, and 82.2 ± 13.6 kg; and for females were 17.1 ± 3.7 years (range 12 to 31 years), 169.2 ± 6.1 cm, and 64.8 ± 9.6 kg. All participants were involved in physical activity (34% participated in netball, 19% in rugby, 9% in field hockey, 9% in soccer, and 29% in other sports). On average, participants were involved in physical activity four times per week, 6 h a week. Participants had to be free from injury, pain, or any other issue that would limit physical activity participation. Previous injuries were not an exclusion criterion. Participants were recruited via word-of-mouth, research contacts, social media, and emails sent to local sports clubs. The study protocol was approved by our institution's health research ethics committee [HREC(Health)#41] and adhered to the Declaration of Helsinki. All participants and their legal guardian when younger than 16 years of age signed a written informed consent document that explained the potential risks associated with testing prior to participation.

#### *2.2. Data Collection*

We used the original LESS protocol for testing [10]. Participants jumped horizontally from a 30 cm high box to a line placed at 50% of their body height, and immediately jumped upward for maximal vertical height. We placed an emphasis on jumping off the box with both feet, landing in front of the designated line, jumping as high as possible straight up in the air once they landed from the box, and completing the task in a fluid motion. We did not provide any feedback on participants landing technique unless they were performing the task incorrectly. Participants used their own footwear for testing.

After task instructions and practice jumps for familiarization (typically 1), each participant performed three successful trials of the double-leg drop-jump landing task in front of two standard video cameras capturing at 120 Hz (Sony RX10 II, Sony Corporation, Tokyo, Japan) with an actual focal length of 8.8 to 73.3 mm (35 mm equivalent focal length of 24–200 mm). We mounted the cameras on tripods placed 3.5 m in front of and to the right side of the landing area with a lens-to-floor distance of 1.3 m. We allowed participants to rest until they felt ready to perform the task again to limit fatigue between the three trials. Total testing time was typically 2 min per participant.

#### *2.3. Clinical LESS*

A qualified physiotherapist who completed over 400 LESS evaluations (IH) replayed the videos using the Kinovea software (version 0.8.15, www.kinovea.org), identified the two key frames of initial ground contact (IC) and maximal knee flexion (KFmax), and scored all trials using the 17-item LESS scoring sheet (Table 1). The clinician was blinded to the results from the automated computer-vision scoring. A total of 320 double-leg drop-jump landings from the potential 432 trials (3 jumps × 144 participants) were retained for analysis because of certain participants not completing three trials, one or both video files being not usable, or a clear misidentification of time events from the automatic cropper described in the following subsection (i.e., more than 100 ms difference with the clinician).

#### *2.4. Automated LESS*

The LESS score prediction algorithm we developed was a multistage process. Generally, the first stage consisted of processing the videos to detect the IC and KFmax key frames, which involved running the frontal and lateral videos for each jump through OpenPose v.1.21 [16], and then using a heuristic method to identify the key frames. Once that stage was complete, we extracted measurements from the key frames to use as features for machine learning. The final stage was the score prediction for the drop-jump landing trial from the features using a machine learning algorithm. The entire process is depicted in Figure 1. We further evaluated the predictive accuracy of the final machine learning stage using cross validation.

**Figure 1.** Flow diagram of data processing leading to comparing 'gold standard' clinical LESS scores from an expert rater to 'automated' predicted LESS scores from the automation process. *Abbreviations*: IC, initial contact; KFmax, maximal knee flexion; LESS, Landing Error Scoring System; RF, random forest.

In more detail, the algorithm used to detect key frames in the first stage is described in Table 2. The input to the algorithm are the frontal and lateral videos for a single drop-jump landing trial, and the output are cropped versions of the same videos where the first and last frames correspond to the IC and KFmax key frames, respectively.

**Figure 2.** This figure is an example of the original (blue line) plot and rolling window (orange line) plot for the right ankle keypoint of one individual during a drop-jump landing trial taken from the lateral view video. More specifically, (**a**) the blue line depicts the distance of the right ankle to the left boarder (y-axis) in each video frame (x-axis); (**b**) the orange line is the 20-frame rolling median of the original blue line; (**c**) the black bars indicate the intersections of the two lines, whereas the red dotted line represents the distance between two consecutive intersection points. Figure (**d**) is a zoomed-in view of the intersections around the initial contact key frame. Figure (**e**) highlights the points (f) and (g) as the initial contact key frame on the rolling window plot and original plot, respectively.

The basic method is to track the location of the ankles (using OpenPose and COCO 18-points model [16]) across the frames to detect the frame in which landing occurs based on the original and rolling window plots (Figure 2), and additionally to track the body and knee keypoints so that the ankle/knee/body angle can be calculated and used to identify the point of maximum knee flexion. Once these two points are identified in both videos, then the frames before and after the key frames are cropped away. This stage generally reduces the length of the original videos from several seconds down to less than 250 ms.

Once cropping is complete, two videos in which the first frame corresponds to IC and the last frame corresponds to KFmax pass to the second stage. In the second stage of processing, features are extracted from both videos and merged into a single 'example' that will be used for machine learning. A total of 42 kinematic features from the two key frames in each video were generated. The features are a mixture of angles between specific OpenPose keypoints (shown in Figure 3) and ratio between distances. The specific features are listed in Table 3. A total of six angles were extracted from all four key frames with an additional eight features (mixture of angles, distances, and distance ratios) being extracted from the two frontal key frames only, for a total of 40 measurements. Two further features, being the length in frames of the cropped frontal and lateral videos, were also included.


**Table 2.** Algorithm used to detect key frames from the two input videos.

*Notes*. <sup>a</sup> Rolling window plot, plot of median values from a rolling 20-frame window. See Figure 2. *Abbreviations.* F, Frontal view video; IC, initial contact; L, Lateral view video KFmax, maximal knee flexion.

**Figure 3.** OpenPose's COCO 18-points model keypoint positions (left image) [16] and example of a frontal (middle image) and lateral (right image) view processed video at the maximal knee flexion key frame.


**Table 3.** Measurements extracted from key frames and used as kinematic features.

*Notes.* Key frames are: (i) initial contact, (ii) maximal knee flexion. <sup>a</sup> Refer to Figure 3 for keypoints.

Following feature extraction, we then used a machine learning algorithm to predict the LESS score associated with the drop-jump landing videos. To evaluate the predictive effectiveness of the various machine learning algorithms, we generated features for all 320 drop-jump landings in the dataset using the approach described above. It was also noticed that the distribution of the LESS scores in the dataset was imbalanced, with the majority of LESS scores falling in the range 4–6. Given that unbalanced datasets can potentially affect the accuracy of machine learning techniques, we additionally generated a balanced version of the dataset consisting of 153 drop-jump landing trials with at most 20 trials per LESS score. All evaluations of machine learning techniques were applied to both datasets.

The machine learning techniques chosen to be evaluated were random forest regression, because it is a state-of-the-art machine learning approach and generally performs well 'out of the box' on most problems in practice; and linear regression, which is a widely understood linear modelling technique. Unlike random forest regression, linear regression produces an interpretable model, but it has the disadvantage of being unable to model interactions between features. Given that the full dataset was imbalanced, we also evaluated a dummy regressor (ZeroR) that simply predicts the mean LESS score from the training data. For the original dataset, this method was expected to have reasonably high accuracy, but lower accuracy for the balanced dataset. All machine learning methods implemented were available in WEKA 3.8.0 [19], and returned floating point numbers (i.e., decimals) that added granularity to the data.

#### *2.5. Statistical Method*

As noted in Section 2.3, 320 double-leg drop-jump landings were analysed. A series of 10 × 10-fold cross validation experiments were applied on full (320 videos) and balanced (153 videos, ≤ 20 videos per LESS score) to predict the scores using random forest for regression, linear regression, and dummy regression (ZeroR) models in WEKA [20]. To assess the effectiveness of the automated cropping algorithm in the context of the overall system, we additionally ran the entire pipeline with crops generated by the clinician. Mean absolute error and Pearson correlation coefficient (*r*) were calculated to assess the accuracy of the predictions. Predictions were then converted to a binary category and sensitivity-specificity for categorising individuals at high risk of non-contact ACL injury (LESS ≥ 5 errors [12]) were assessed for each method. The outcomes of the models were compared using paired corrected *t*-tests in WEKA [20], and the timestamps of the key frames IC and KFmax respectively compared between manual (clinician) and automated (OpenPose) cropping methods using unpaired *t*-tests assuming homoscedasticity. Since the LESS score was treated as a regression problem, actual (clinical LESS) versus predicted (automated LESS) and Bland-Altman [21] plots were used to allow for a visual inspection of the models. Statistical significance was set at *p* ≤ 0.05.

#### **3. Results**

The mean LESS score from the 320 drop-jump landings was 5.5 ± 1.8 errors (range 0 to 12 errors) as rated by the clinician. The absolute time difference between manually identified IC and KFmax was 26.5 ± 17.0 (*p* = 0.484) and 32.8 ± 18.0 ms (*p* = 0.445) for the frontal videos, and 53.5 ± 16.2 (*p* = 0.125) and 20.8 ± 16.3 ms (*p* = 0.827) for the sagittal videos.

Random forest yielded the lowest mean absolute error (1.23) and greatest correlation (*r* = 0.63) between actual and predicted scores based on results from the cross validation experiments (Table 4). Sensitivity (0.82) and specificity (0.77) were reasonable for high (LESS ≥ 5 errors) and low (LESS < 5 errors) injury risk categorisation. Experiments using a balanced (versus unbalanced) dataset or manually (versus automated) cropping methods did not improve predictions. An actual versus predicted plot from the random forest regression is depicted in Figure 4, and two Bland-Altman plots on the same dataset in Figure 5. Note that both conventional (mean difference ± 1.96 standard deviation) and regression-based (regressed difference between methods on the mean of the two methods ± 2.46 standard deviation of the residual) Bland-Altman plots were generated given the non-uniform differences in mean [21].



*Notes*. Values are means ± standard deviations. *Abbreviations*. RF, random forest. \* Significant difference versus random forest (*<sup>p</sup>* ≤ 0.05) using paired-corrected t-tests. <sup>a</sup> Categorising high (LESS ≥ 5 errors) and low (LESS <sup>&</sup>lt; 5 errors) injury risk individuals [12].

**Figure 4.** Actual (clinical) versus predicted (automated) LESS score plots from the random forest regression using full dataset (n = 320) and automatic cropping method. Dashed lines represent the 5-error threshold that defines high risk of injury (i.e., scoring 5 or more errors during LESS has been associated with a 10.7 times greater relative risk of sustaining a non-contact anterior cruciate ligament injury [12]). Note that the clinical scores are integers and predicted scores are decimals, which adds granularity. *Abbreviations*: LESS, Landing Error Scoring System.

**Figure 5.** Bland-Altman [21] plots depicting the difference in predicted (automated) and actual (clinical) LESS scores versus the mean scores with (**A**) conventional 95% limits of agreement (mean difference ± 1.96 standard deviation), and (**B**) regression-based limits of agreement (regressed difference between methods on the mean of the two methods ± 2.46 standard deviation of the residual).

#### **4. Discussion**

The use of the LESS to assess injury risk is common in sport science and clinical practice [9,22], but scoring is time consuming, clinician-dependent, and generally inaccessible for large-scale screening outside of elite sports. This study provides evidence that the LESS can be automated using deep-learning-based computer vision combined with machine learning methods without the need for 3D motion analysis or depth sensor cameras. A clear benefit of automating LESS scoring is immediate feedback to end-users. The successful application of this method paves the way for the automatic detection of individuals at high risk of injury using smartphone-based applications of LESS videos (Video S1: https://youtu.be/q1wiGt4K8MU).

The characteristics of an ideal injury risk screening tool are good reliability, validity, and predictive value for injury incidence. In practical or field settings, an ideal screening method is easy to administer without an expert, and has minimal financial, spatial, and temporal requirements. Ideally, the screening tool provides immediate results and is accessible to everyone, from the recreational to elite athlete, as well as novice to expert rater. Overall, the LESS responds to most of these stated requirements. The test demonstrates acceptable reliability and validity [10,11,23], as well as predictive value for non-contact ACL injury using a threshold of 5 errors [12]. The inter-rater reliability of the total LESS score is good to excellent, with ICC ranging from 0.83 to 0.92 [10,11,23] and typical errors at 0.71 LESS errors [10]. The results from the current study indicate that the typical errors from the automated processing and scoring of the LESS through computer vision when applying the random forest model (Table 2) are less than half an error greater than scores taken from two expert clinicians. In fact, certain individual LESS items yield suboptimal psychometric properties between raters and 3D motion analysis [23]. More specifically, no significant agreement between raters was found for knee and trunk flexion at IC, and poor agreement between rater and 3D motion capture analysis was found for knee flexion at IC, lateral trunk flexion at IC, and symmetric foot contact at IC [23]. As such, a certain level of disagreement between clinical ratings and computerised ratings is expected.

As seen in Figures 4 and 5, the estimated error is not uniform across the range of LESS scores, but depends on the target. For example, trials with a low actual LESS score tend to have a positive error (the prediction is an overestimation) and trials with a higher actual LESS score tend to have a negative error (the prediction is an underestimation). If these biases stemmed from the over representation of the mid-range LESS values (i.e., majority of LESS scores falling in the range 4–6), the balanced dataset should have provided more accurate predictions, which was not the case. It might be possible to attempt correcting predictions to improve accuracy in future work using probability calibration methods, such as Platt Scaling and Isotonic Regression. The large errors in LESS score predictions were attributed to inaccurate foot and IC key frame detection. The newest body model in OpenPose (Body 25) contains 25 points, including coordinates that define the feet and enable computations of angles at the ankles [24]. Improving the LESS score automation relies on either refining body part detection or training a new system specifically to solve this problem.

In previous research, depth sensor technology has been used to automate LESS scoring [13,15]. Comparisons between automated and expert clinicians indicate a mean difference of 1.20 errors [15], mean absolute difference of 1.13 errors [13], intra-class correlation of 0.80 [16], and percentage agreement of the individual items ranging from 55–100% [13,15]. These research findings are comparable to our lowest mean absolute error (1.23), greatest correlation (*r* = 0.63), and agreement in risk classification (sensitivity 0.82, specificity 0.77) between actual and predicted scores from the cross validation experiments using random forest regression. In contrast to the PhysiMax system [13,15], our approach did not require the clinician to add the overall impression manually (no. 17 in Table 1) given that the LESS items were not scored one-by-one. Although the lack of individual-item scores might be perceived as a limitation of the deep LESS approach; no subjective rating from the clinician or hardware other than a handheld camera or smart portable device are required. Furthermore, only the final LESS score has shown predictive value in terms of injury risk [12]; hence, the individual items are of lesser clinical value.

The better accuracy achieved by random forest can be explained by the fact that the features (angles, distances, and ratios) are likely correlated and related in a non-linear manner. Decision tree ensembles in general are better able to cope with correlated variables and model non-linear patterns [25]. Linear regression, on the other hand, achieves optimal results when the predictor variables are independent and do not interact. We also foresee a possibility of processing the raw video images themselves and attempting direct deep learning-based classification with minimal pre-processing. Such an approach would obviate the need to use OpenPose or a similar pose-tracking tool. However, taking such an approach would be challenging because of the lack of training data relative to size of datasets usually used to train deep image recognisers. Another significant disadvantage of the proposed approach is that deep learning needs GPU-based acceleration hardware, and is therefore currently unable to process videos independently on consumer smartphones. That said; the rapidly increasing computational power of consumer smartphones and the current trend in research of compressing deep models [26] so that they run efficiently on mobile devices should solve this problem in the next few years.

One of the main concerns in clinical screening tools are their subjective nature and reliance on visual observations to estimate angles, which are challenging to quantify accurately [27,28]. During the LESS, a small kinematic difference (e.g., knee angle 29◦, 1—error present; knee angle 30◦, 0—error absent) can result in poor agreement between raters and between clinical LESS scores and motion capture scores. Recent technological advances have allowed the more objective quantification of human motion using wearable technology [29,30]. Inertial measurement units are able to measure linear and angular motion of individual body segments and centre of mass, and are proposed as more accurate means of identifying risky movement patterns than through visual observations [31]. Although inertial measurement units are relatively inexpensive; they are not commonly used in clinical environments and an expert is still needed to process and interpret data signals. The automated scoring process here developed using standard video recordings offers an alternative solution that can possibly improve consistency of LESS ratings, removing the subjective interpretation of the task. Moving forward, the reliability of deep LESS scores, validity of OpenPose derived data during the dynamic double-leg drop-landing task, and predictive ability of the method need empirical support.

An indisputable advantage of automated scoring using deep-learning-based computer vision combined with machine learning methods or markerless methods from depth sensor cameras is immediate results and feedback to patients, athletes, coaches, or healthcare professionals. Our developed method that automates LESS scores provides a viable solution to decreasing scoring time, increasing accessibility to non-expert raters, and delivering immediate results without any additional expenditure other than conventional video recordings. Conventional 2D video recordings are adequate for quantifying kinematics [32–34] and are readily accessible through tablets or smartphones. The successful application of this method would pave the way for the automatic detection of individuals at high risk of injury using smartphone-based applications of LESS and 2D video footage (Video S1: https://youtu.be/q1wiGt4K8MU). Other than expediting mass injury risk screening initiatives in youth or team sports, LESS automation could be a valuable and convenient tool to track injury risk factors over time and to assess the effectiveness of intervention programs at improving landing mechanics (Video S2: https://youtu.be/Ve\_QJu0fuLs). The proposed method could be extended to other injury risk screening methods based on 2D camera recordings to decrease manual labour and time required for screening initiatives; e.g., the Cutting Movement Assessment Scale [35] and Tuck jump assessment [36].

This preliminary investigation provides evidence that it is feasible to automate the LESS from 2D video recordings alone. Further research could lead to improved automation outcomes and enhance the strength of the agreement between clinical and automated LESS scores beyond its current levels. The newest body model in OpenPose (Body 25) contains 25 points, including coordinates that define the feet and enable computations of angles at the ankles [24]. Although the timestamped IC key frame in frontal and sagittal videos were comparable between the clinician and scripted process (mean difference: 32.8 ms, *p* = 0.445 and 20.8 ± 16.3 ms, *p* = 0.827), using the foot coordinates rather than ankle and body coordinates would certainly enhance precision. A number of videos from the available dataset were not used because of a clear misidentification of time events from the automatic cropper (i.e., more than 100 ms). We were unable to determine the reason underlying the mislabelling of these videos upon visual inspection. We speculate that rerunning the current experiment using the COCO + Foot model might lead to the correct identification of key events in a greater number of our database videos, increasing the number of eligible videos for analysis. The increased number of coordinates from the 25-point Body rather than 18-point COCO model would also allow us to extract a greater number of features from the processed videos and use these as input in the subsequent regression experiments.

#### **5. Conclusions**

We provide evidence that the Landing Error Scoring System (LESS)—an injury-risk screening tool—can be automated using deep-learning-based computer vision combined with machine learning methods. Further research on the automation would enhance the strength of the agreement between clinical (gold standard) and automated (predicted) LESS scores, and risk classification beyond its current levels. Automation of the LESS using standard 2D recordings would facilitate mass injury-risk screening initiatives with quasi real-time feedback, without the need of depth cameras or expert clinicians. The successful application of this method would pave the way for the automatic detection of individuals at high risk of injury using smartphone-based applications of LESS and 2D video footage (Video S1: https://youtu.be/q1wiGt4K8MU), increasing accessibility of injury-risk assessment methods beyond elite athletes and removing depth-sensor camera requirements. It may also open doors to other related injury prevention problems. Future work includes updating the framework using the newest body model in OpenPose (Body 25) to extract a greater number of features and more accurately detect key frames.

**Supplementary Materials:** The following are available online: https://youtu.be/q1wiGt4K8MU. Video S1: LESS demonstration. https://youtu.be/Ve\_QJu0fuLs. Video S2: The 'DEEP' Landing Error Scoring System.

**Author Contributions:** Conceptualization, K.H.-L.; methodology, K.H.-L.; formal analysis, K.H.-L., L.S., M.M.; investigation, K.H.-L., I.H.; data curation, I.H., C.Z.; writing—original draft preparation, K.H.-L., I.H.; writing—review and editing, K.H.-L., I.H., C.Z., L.S., M.M.; supervision, K.H.-L., M.M.; project administration, K.H.-L.; funding acquisition, K.H.-L., L.S., M.M. Authorship must be limited to those who have contributed substantially to the work reported. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by a University of Waikato Strategic Investment Fund 2018 Medium research Grant.

**Acknowledgments:** We would like to acknowledge Ruili Wang for expert advice and Christopher Martyn Beaven for research support.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

#### *Article*
