EMG Based Pattern Recognition and Classification in Post Stroke Patients

This project deals with testing advanced dimensions-reduction algorithms on data, which was recorded from EMG signals of healthy subjects and post stroke patients. The project has two parts - Signal processing: noise reduction, finding electrical activity parts of the signal etc. and Machine learning part: features extraction, dimensions reduction, training and classification.
During the project, our main goal was to classify EMG signals based on movement directions of the subjects. I.e. the subjects pointed on a specific target and our goal was to predict the movement direction based on the other movements of that subject. Another mission was to examine whether it is possible to classify the stroke severity of the patient by its EMG signals. In addition, we wanted to compare different approaches for features extraction from the data and to test its suitability to our problem.
The two main approaches for features extraction we used were comparing between analyzing the shape of the signal with the connection between the muscles of the same movement. For our data, we got better results in classification of movement directions when analyzing the shape of the signal. However, even though we got good results when classifying movements, we could not classify the post stroke patients' by its stroke severity, probably because of lack of data.

EMG Based Pattern Recognition and Classification in Post Stroke Patients
EMG Based Pattern Recognition and Classification in Post Stroke Patients
EMG Based Pattern Recognition and Classification in Post Stroke Patients
Collaboration:

Dr. Sharon Israeli, Haifa University