AI-based Identification of Arm Movements for Post-Stroke Rehabilitation

Department:  Cognitive Robotics (CoR), Delft Centre for System and Control (DCSC)

Daily supervisor: Sahel Akbari (Erasmus MC/TU Delft) – s.akbari@erasmusmc.nl

Co-supervisors: Arkady Zgonnikov – a.zgonnikov@tudelft.nl, M. (Manon) Kok, M.Kok-1@tudelft.nl

Keywords: Machine Learning, Movement Identification, Stroke Rehabilitation

Background:

Wearable sensors have become valuable tools for therapists to monitor and provide feedback on the arm and hand movements of stroke patients. Recent advances in machine learning and deep learning, combined with data from Inertial Measurement Unit (IMU) sensors, have significantly improved our ability to track and analyze specific movements. The degree of arm recovery following a stroke is closely associated with the extent to which the affected arm is actively used in daily activities. Consequently, there is an increasing focus on monitoring arm movements in less structured, real-world scenarios where movement quality may be lower. However, practically, home-based rehabilitation tools for accurately identifying and assessing arm movements in these unstructured settings remain scarce.

Objective:

In this study, your objective will be to develop novel AI-based techniques to automate detection of different types of arm movements of stroke patients. You will have access to several arm movement datasets captured using IMU sensors from both healthy participants and post-stroke patients. These datasets provide rich information on arm movements collected with different level of control, freedom, and instructions provided to the participants. Using these datasets, you will explore various AI-based approaches with the ultimate goal to develop a system that can generalize the identification of arm movements from controlled, structured settings to real-life environments, enabling the next generation of at-home rehabilitation solutions. Your tasks may include (but not limited to):

• Conducting a literature review on current available AI techniques for movement recognition in people with neurological conditions in their daily life and techniques for automatic movement annotations.
• Designing and implementing an AI-based solution, potentially based on existing open-source solutions for classification of arm movements.
• Evaluating the proposed model(s) using the available data from stroke patients and assessing the model’s accuracy, sensitivity, and ability to generalize across different environments
• Investigating additional data features (such as sensor kinematics outcomes) that could improve classification accuracy.

Requirements:

• Basic Python programming skills
• Familiarity with one of the following (optional, but possible to learn during the assignment too): machine learning, human movement data processing, any kind of signal processing

Contact:

To indicate your interest in the project, email s.akbari@erasmusmc.nl with your CV and academic transcript.

References:

[1]- Biswas, Dwaipayan, et al. “Recognizing upper limb movements with wrist worn inertial sensors using k-means clustering classification.” Human movement science 40 (2015): 59-76.

[2]- Kaku, Aakash, et al. “Towards data-driven stroke rehabilitation via wearable sensors and deep learning.” Machine Learning for Healthcare Conference. PMLR, 2020.

[3]- Panwar, Madhuri, et al. “Rehab-net: Deep learning framework for arm movement classification using wearable sensors for stroke rehabilitation.” IEEE Transactions on Biomedical Engineering 66.11 (2019): 3026-3037.

[4]- Chae, Sang Hoon, et al. “Development and clinical evaluation of a web-based upper limb home rehabilitation system using a smartwatch and machine learning model for chronic stroke survivors: prospective comparative study.” JMIR mHealth and uHealth 8.7 (2020): e17216.

[5]- Chen, Yuqing, and Yang Xue. “A deep learning approach to human activity recognition based on single accelerometer.” 2015 IEEE international conference on systems, man, and cybernetics. IEEE, 2015.

[6]- Panwar, Madhuri, et al. “CNN based approach for activity recognition using a wrist-worn accelerometer.” 2017 39th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC). IEEE, 2017.

[7]- Murad, Abdulmajid, and Jae-Young Pyun. “Deep recurrent neural networks for human activity recognition.” Sensors 17.11 (2017): 2556..

[8]- Li, Min, et al. “A CNN-LSTM model for six human ankle movements classification on different loads.” Frontiers in Human Neuroscience 17 (2023): 1101938.