1. Introduction
Rehabilitation assessment is conducted to determine the nature, location, severity, development trend, scope, and prognosis of dysfunction in patients with disabilities [
1,
2]. It can reflect the situation of dysfunction in patients and lay a scientific foundation for the formulation and implementation of a rehabilitation treatment plan [
3,
4]. Therefore, it is necessary to conduct an objective, accurate, and convenient assessment in the process of rehabilitation training.
In the field of rehabilitation medicine, the focus of rehabilitation assessment is the assessment of limb motion function [
5,
6]. At present, the assessment method is usually evaluated by therapists using a clinical assessment scale, including the are Brunnstrom assessment method, Fugl-Meyer assessment (FMA) scale, and Barthel index [
7,
8]. The FMA method is recognized as one of the most widely used rehabilitation assessment methods. The FMA method has the advantages of detailed content, as well as high assessment reliability and sensitivity [
9], but a single assessment of patients takes a long time, and it is mainly subjectively evaluated by rehabilitation doctors, which cannot ensure objective and unified results [
10].
Robot-derived measurement technology adds a new dimension to the assessment of motion function [
11]. Through the sensor hardware and control system software on the rehabilitation robot, quantitative analysis can be realized, rehabilitation efficiency can be improved, cost can be reduced, and an accurate, objective, and real-time rehabilitation assessment method has become a possibility and a trend [
12,
13].
In recent years, it has become an active topic to seek more excellent rehabilitation assessment methods based on rehabilitation robots [
14,
15]. Some assessment methods have been developed. Wu [
16] designed two tasks of “following the circle” and “crossing the tunnel” to evaluate and classify healthy people and stroke patients in Brunnstrom phase VI using a neural network, but this method is more suitable for patients at the later stage of rehabilitation. Kurillo [
17] predicted the three-dimensional (3D) reachable spatial surface area using Kinect-captured data to evaluate the upper-limb function of patients with facial shoulder brachial muscular dystrophy. Bai [
18] collected the upper-limb movement information through Kinect, calculated the upper-limb reachable space, and used the adaptive network fuzzy inference system to evaluate the patient’s upper-limb rehabilitation training results; however, once the camera of the motion capture system in the above two methods is calibrated successfully, it cannot be moved casually; thus, they are not portable and efficient. Gao [
19] studied a three-degree-of-freedom upper-limb exoskeleton real-time rehabilitation training system based on a surface electromyography (sEMG) signal. After collecting the sEMG signal, four elbow movements can be identified and evaluated by decision tree algorithm. Compared with the traditional rehabilitation assessment methods, this method has more specific assessment results, but relying only on the sEMG signal which is easil disturbed can lead to poor stability. Su [
20] combined sEMG and inertia information with clinical rehabilitation assessment indices to realize a comprehensive and quantitative upper-limb rehabilitation state assessment and database system. However, the assessment system has a small number of test patients and needs to be further improved for clinical practice. Antonella [
21] proposed a multi-parameter method to evaluate the rehabilitation of patients, including sEMG, electroencephalogram (EEG), kinematics, and clinical scale. This multisource method can better characterize the rehabilitation of patients, but it is highly complex and requires the assistance of professionals. All these measurement methods have their own advantages; however, at present, most robot rehabilitation assessment technologies still need the assistance of professionals, and there is a lack of large samples and high-quality long-term clinical studies proving their accuracy and reliability [
22,
23].
Generally speaking, there are no perfect and standardized assessment methods, and this also applies to the rehabilitation assessment of finger function [
24]. Using the innovative design of the End-Effector Finger Rehabilitation Robot (EFRR), combined with multisource information such as finger contact pressure, sEMG signal, and finger joint range of motion, this paper formulates a comprehensive rehabilitation assessment method suitable for this system, which has the characteristics of miniaturization and home use without a physician’s assistance. It provides a new idea and possibility for the objective and real-time automatic rehabilitation assessment of finger motion function.
4. Discussion
As shown in
Figure 11, the pressure of the ring finger, middle finger, and little finger is high in the process of rehabilitation; the middle finger and ring finger are commonly used for force, whereas the contact between the pulp of the little finger and the finger-cot is not completely horizontal in the process of flexion. Therefore, there is a systematic error caused by the physical displacement of the little fingertip. At the same time, the reason for the small force of the index finger is that the finger-cot space is fixed and cannot fully fit each finger. Therefore, the gap between the index finger and the finger-cot is large, resulting in failure of the index finger to directly touch the pressure sensor throughout the rehabilitation process. According to the above problems, we can see that there were some defects in the finger-cot design of the rehabilitation robot, leading to certain errors in fingertip pressure detection during the rehabilitation process; this will be addressed in future generations of the robot for mechanism improvement.
Figure 18 and
Figure 19 compare the eigenvalues of the three volunteers, where it can be seen that both MAV and RMS had a significant weakening trend before and after training. This shows that the effect was not caused by individual differences, whereas the embodiment of individual differences was mainly reflected in the size of their eigenvalues. The purpose of active training was to get the muscles to a state of fatigue, thereby highlighting the relationship between muscle fatigue and the MAV and RMS. Therefore, for the judgment of muscle fatigue, in addition to the subjective assessment of people, we can analyze their physiological signal data.
This paper designed a finger muscle strength estimation model based on fingertip pressure, a fatigue monitoring system based on sEMG signal, and a joint ROM estimation model based on a motor encoder, which provides a novel assessment method based on multi-sensor data. Compared with the existing common assessment methods, the comprehensive rehabilitation assessment method proposed in this paper can be completed automatically through multi-source data collection to help patients accurately locate their own rehabilitation situations, which has lower requirements for experienced rehabilitation doctors, and the collection method is simple and efficient. This has positive significance for realizing home treatment, thereby alleviating the shortage of rehabilitation doctors, reducing costs, and improving the efficiency of rehabilitation assessment.
5. Conclusions
In order to realize the objective, accurate, and convenient assessment of finger rehabilitation effect, a new finger rehabilitation assessment method was designed using the developed End-Effector Finger Rehabilitation Robot, which integrates muscle strength, fatigue degree, and joint ROM. Firstly, a finger muscle strength estimation model based on fingertip pressure was established, and the estimated muscle strength was scored and graded using the entropy weight method. Then, a fatigue monitoring system based on sEMG signal was designed to determine the fatigue degree of people in the training process by collecting the sEMG signals of muscles. Lastly, combined with the finger-joint ROM indicator, a comprehensive rehabilitation assessment model of finger function was established by using the AHP assessment method, which can achieve real-time assessment using multi-sensor signals in the process of rehabilitation training. The effectiveness of the rehabilitation assessment method was verified by experiment, providing the possibility of helping patients automatically, accurately locating the rehabilitation situations, and realizing convenient and objective finger rehabilitation assessment.
In the future, after optimizing the mechanical system and control system, the EFRR will be applied to real patient samples instead of healthy volunteers to further verify the feasibility of the comprehensive assessment method of finger function. Using the results of this comprehensive rehabilitation assessment method, an online system will be designed to realize home-based independent rehabilitation training and real-time rehabilitation assessment of patients.