**1. Introduction**

Inverse dynamics analysis is a fundamental tool widely used for biomechanical studies to understand human movement. The inverse dynamics method combines kinematic and kinetic data with anthropometric parameters and can estimate joint moments and powers [1]. The evaluation of joint moments and powers is critical in clinical decision-making, such as gait retraining [2], treatment with insoles or orthoses [3], and even surgery [4]. Despite its widespread use, the inaccuracy of inverse dynamic analysis stemming from kinematic/kinetic/anthropometric data is well-recognized [5]. Currently, most kinematic data are provided by marker-based (MB) motion-capture systems [6]. However, inaccuracies derived from marker placements, including the center of mass locations [7,8], joint centers [9], the noise due to surface marker movement [10], and skin artifacts [11,12], can be significant barriers. In addition, using MB requires highly trained personnel to avoid human errors when placing markers on participants [13,14]. The intensive time commitment for marker placements and a controlled environment [15] also contribute to the drawbacks of MB. These all make the applications of MB systems challenging in clinical settings with clinical populations [16].

**Citation:** Tang, H.; Pan, J.; Munkasy, B.; Duffy, K.; Li, L. Comparison of Lower Extremity Joint Moment and Power Estimated by Markerless and Marker-Based Systems during Treadmill Running. *Bioengineering* **2022**, *9*, 574. https://doi.org/ 10.3390/bioengineering9100574

Academic Editors: Christina Zong-Hao Ma, Zhengrong Li and Chen He

Received: 9 September 2022 Accepted: 11 October 2022 Published: 19 October 2022

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2022 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 (https:// creativecommons.org/licenses/by/ 4.0/).

As a critical advancement in vision-based motion capture, a markerless (ML) system offers an alternative approach to measuring kinematic data. Studies have shown the applicability of the deep learning algorithm-based markerless system in gait analysis. Kanko et al. reported excellent agreement with the MB system on spatial parameters (e.g., step length, stride width, and gait speed) and slight differences in temporal parameters (swing time and double support time) [17]. In two follow-up studies, they assessed the lower extremity joint center positions and joint angles. One study emphasized the intersession variability of joint angles. They reported that the average inter-session variability across all joint angles was 2.8◦ in the ML system, which is less than all previously reported values (3.0–3.6◦ ) for the MB system [18]. Their other study presented the average systematic root-mean-square joint center differences of 2.5 cm except for the hip joint, which was 3.6 cm. The average systematic root-mean-square for all segment angle differences was 5.5◦ , except for the rotation angle about the longitudinal axis [19]. These strong results are approaching or superior to the accuracy of MB systems. However, we did not find any lower extremity joint moments and powers in comparison between the two types of systems in the literature. Given the fundamental quantities of interest in human motion research are the intersegmental moments acting at the joints [20], it is critical to compare joint moments estimated by the ML and MB systems.

Due to marker-based systems' weakness, markerless systems might introduce new possibilities for inverse dynamic analysis. Therefore, the purpose of the current study was to compare inverse dynamic outcomes of lower extremity joint moments and powers based on ML and MB motion capture systems.

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

#### *2.1. Participants*

Recreationally active young adults were recruited for the current study. Inclusion criteria were: (1) free of musculoskeletal injuries and operations of the lower extremity at least 6 months before the data collection and (2) experience with treadmill running. Participants were asked to be free from any intensive exercise within the 24 h before data collection. Participants signed consent forms approved by the Ethics Committee of Georgia Southern University (Approval Number: H22327) before data collection.

#### *2.2. Experimental Setup and Procedure*

Two camera systems were used in the motion capture procedure: 8 infrared cameras (Vicon Bonita, Oxford, UK) for the MB system to record marker trajectories; and 8 highresolution video cameras (Vicon Vue, Oxford, UK) for the ML system to record movements. The resolutions of Bonita and Vue cameras are 1 megapixel (1024 × 1024) and 2.1 megapixels (1920 × 1080). Cameras were aimed at the instrumented treadmill (AMTI force-sensing tandem treadmill, Watertown, MA, USA) within a 15.5 m long by 7.6 m wide by 2.4 m tall laboratory space. Camera systems and the instrumented treadmill were synchronized using Vicon Lock+ (Vicon, Oxford, UK), where kinematics were recorded at 100 Hz, and the ground reaction forces were recorded at 1000 Hz.

Before data collection, the cameras were calibrated using an Active Wand (V1, Oxford, UK). Calibration for the Bonita cameras of the MB system and the Vue cameras of the ML system included more than 1000 frames of valid wand data and 600 frames of valid wand data, respectively. The tolerance of image error for MB and ML systems was set as 0.2 and 0.4, respectively. The three-marker option, with an origin marker and two markers for the *X*- and *Y*-axis, was used to set the MB system's global coordinate system (GCS). An MTD-3 device and CalTester software (CalTester, Motion Lab Systems Inc., Baton Rouge, LA, USA) were used to examine the spatial synchronization between the force plates and cameras, following the manufacturer's recommended protocol.

When participants arrived at the laboratory, each was introduced to the test protocol. Then, each participant changed into tight shirts and shorts provided by the lab and wore their running shoes. The investigator measured their heights and body mass. Five-minute

warm-up exercise and familiarization with treadmill running followed. Following the manufacturer's suggested procedure, twenty-six 14 mm retro-reflective markers were attached to the participant's anterior superior iliac spine, posterior superior iliac spine, most lateral prominence of the greater trochanter, lateral prominence of the lateral femoral epicondyle, medial prominence of the medial femoral epicondyle, proximal tip of the head of the fibula, anterior border of the tibial tuberosity, lateral prominence of the lateral malleolus, medial prominence of the medial malleolus, dorsal margin of the first, second and fifth metatarsal head, and aspect of the Achilles tendon insertion on the calcaneus at both sides. A static trial for the MB system was recorded in advance while the participants stood on the treadmill with an anatomical posture. Each participant started walking at an initial speed of 1.12 m/s and then gradually transitioned to running at the target speed of 3.58 m/s. The participants ran on the treadmill for 120 s at 3.58 m/s, and the last 30 s of running were recorded for further analysis.
