Aleksandar (Alex) Vakanski

 

Human Movement Modeling in Biomedical Applications

Physical therapy and rehabilitation is essential for functional recovery after stroke, surgery, or musculoskeletal trauma. Although the largest portion of rehabilitation is performed in a home-based setting, numerous barriers lead to low patient motivation and poor adherence rates to the prescribed rehabilitation plans. The primary barrier impacting rehabilitation compliance is the lack of continuous monitoring, supervision, and feedback of patient performance and progress by a healthcare professional. Consequently, the critical need for the design and development of tools to support home-based therapy in the outpatient setting has been widely recognized. The absence of such tools will continue to negatively impact the length of rehabilitation programs and healthcare costs.

The project aims to exploit recent progress in deep artificial neural networks (NNs) toward the development of methods for modeling and assessment of human motions in physical rehabilitation applications. Motivation for this project stems from the demonstrated potential of deep NN architectures to encapsulate highly nonlinear relations among sets of observed and latent variables, as well as the capacity to encode data features at multiple hierarchical levels of abstraction. These properties enabled deep NNs to outperform conventional machine learning approaches across various tasks and application domains.

Although most prior works on automated assessment of movement quality in rehabilitation programs employ distance-based functions for movement evaluation, developing effective and practical approaches is strongly predicated on the provision of effective and powerful models of human movements. Hence, the abilities of deep NNs for hierarchical data representation could provide important advantages in modeling spatial and temporal variations in movement data. On the other hand, training deep NN models requires large databases of labeled examples, and yet, creating databases is a painstaking, expensive, and time-consuming process. This project aims to close this loop by: (a) formulating metrics for assigning clinical quality scores to rehabilitation movements; (b) creating open datasets of rehabilitation exercises; and (c) designing deep NNs for movement modeling and assessment.

Our long-term goal is to develop a commercial system for assessment of patient performance in home-based rehabilitation programs. The system will employ: (1) a color/depth sensor (e.g., a Kinect like camera) for capturing a patient’s movements during sessions, (2) a personal computer, and (3) a computer software application for data analysis, end-user interaction, and data streaming to a healthcare provider. Providing that a patient already possesses a personal computer, the system will be inexpensive with the cost encompassing the color-depth sensor (sold for $150-$250) and the software program. An alternative solution will be to simply use a cell-phone instead, provided that the motion capturing abilities of cell-phone cameras will achieve sufficient accuracy for rehabilitation assessment.

To this end, we conducted preliminary studies toward automated assessment of rehabilitation exercises, based on funding support of an NIH-CMCI (Center for Modeling Complex Interactions at the University of Idaho) grant. Our prior work includes a framework for movement assessment using a spatio-temporal NN, statistical approaches for movement modeling, and a taxonomy of metrics for assessing the consistency of human movements.

We created a dataset UI-PRMD (University of Idaho – Physical Rehabilitation Movement Dataset). The dataset is freely available to the public and involves 10 repetitions of 10 exercises performed by a group of 10 healthy participants. The movements were recorded with a Vicon optical tracking system and a Kinect sensor.

In another work, our team designed a deep learning model based on generative adversarial networks (GAN) for assessment of individual repetitions in rehabilitation exercises. For instance, the right figure below displays the ground truth quality scores for a set of repetitions and the predicted quality scores by the network, where the predicted scores closely follow the true scores. The proposed GAN model was also trained to generate synthetic repetitions that resemble the human captured data, which can potentially be used to augment datasets of rehabilitation movements.

 

Publications

1. Y. Liao, A. Vakanski, M. Xian, D. Paul, and R. Baker, "A review of computational approaches for evaluation of rehabilitation exercises," Computers in Biology and Medicine, vol. 119, article no. 103687, Apr. 2020. [Bibtex] [Elsevier Science Direct

2. Y. Liao, A. Vakanski, and M. Xian, "A deep learning framework for assessing physical rehabilitation exercises," IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 28, no. 2, pp. 468–477, Feb. 2020. [Bibtex] [IEEE Explore

3. C. Williams, A. Vakanski, S. Lee, and D. Paul, "Assessment of physical rehabilitation movements through dimensionality reduction and statistical modeling," Medical Engineering & Physics, vol. 74, pp. 13–22, Dec. 2019. [Bibtex] [Elsevier Science Direct]

4. L. Li, and A. Vakanski, "Generative adversarial networks for generation and classification of physical rehabilitation movement episodes," International Journal of Machine Learning and Computing, vol. 8, no. 5, pp. 428–436, Oct. 2018. [Bibtex] [IJMLC

5. A. Vakanski, H-p. Jun, D. Paul, and R. Baker, "A data set of human body movements for physical rehabilitation exercises," Data, vol. 3, no. 2, pp. 1–15, Jan. 2018. [Bibtex] [MDPI Data in Science

6. A. Vakanski, J. M. Ferguson, and S. Lee, "Metrics for performance evaluation of patient exercises during physical therapy," International Journal of Physical Medicine and Rehabilitation, vol. 5, no. 3, pp. 1–6, Jun. 2017. [Bibtex] [Int. J. Phys. Med. Rehabil.

7. A. Vakanski, J. M. Ferguson, and S. Lee, "Mathematical modeling and evaluation of human motions in physical therapy using mixture density neural networks," Journal of Physiotherapy and Physical Rehabilitation, vol. 1, no. 4, pp. 1–10, Dec. 2016. [Bibtex] [Physiother. Rehabil.