Automated Detection of Rehabilitation Exercise by Stroke Patients Using 3-Layer CNN-LSTM Model

Area of Science:

• Rehabilitation medicine
• Artificial Intelligence
• Human Activity Recognition

Background:

• Stroke is a leading cause of adult disability and mortality, necessitating effective rehabilitation strategies.
• Physiotherapy is crucial for restoring motor function and daily living activities post-stroke.
• Automated detection of rehabilitation exercises presents a significant challenge in human activity recognition.

Purpose of the Study:

• To develop and evaluate an automated approach for detecting various therapy exercises performed by stroke patients during rehabilitation.
• To leverage deep learning techniques for accurate human activity recognition in the context of stroke recovery.

Main Methods:

• A deep learning model, the 3-Layer Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM), was proposed.
• The model utilized RGB camera data, preprocessed through resizing.
• The CNN layers extracted features, followed by fully connected layers and an LSTM layer for spatial-temporal dynamics.

Main Results:

• The 3-Layer CNN-LSTM model achieved an accuracy of 91.3% in detecting rehabilitation exercises.
• Performance was compared against a standalone CNN model (89.9% accuracy) and a K-Nearest Neighbors (KNN) algorithm.
• The proposed 3-Layer CNN-LSTM model demonstrated superior performance over the comparative methods.

Conclusions:

• The 3-Layer CNN-LSTM model offers a promising automated solution for monitoring stroke rehabilitation exercises.
• Accurate exercise detection can support physiotherapists in patient progress assessment and therapy personalization.
• This deep learning approach advances human activity recognition in clinical settings for improved stroke recovery outcomes.