Gradient boosting DD-MLP Net: An ensemble learning model using near-infrared spectroscopy to classify after-stroke...

Area of Science:

• Biomedical Engineering
• Neurorehabilitation
• Machine Learning in Healthcare

Background:

• Stroke survivors often experience dyskinesias, impacting motor function and quality of life.
• Accurate assessment of dyskinesias is crucial for effective rehabilitation planning.
• Current assessment methods can be subjective and time-consuming.

Purpose of the Study:

• To develop an automated system for assessing the degree of post-stroke dyskinesias.
• To combine near-infrared spectroscopy (NIRS) with machine learning for objective evaluation.
• To validate the model's accuracy in classifying dyskinesias across different motor stages.

Main Methods:

• Recruited 35 subjects across five stages (healthy and Brunnstrom stages 3-6).
• Utilized NIRS to capture hemodynamic responses in upper and lower limb muscles during passive and active exercises.
• Employed D-S evidence theory for feature fusion and a Gradient Boosting DD-MLP Net model for dyskinesia evaluation.

Main Results:

• Achieved high classification accuracy for upper limb dyskinesias: 98.91% (passive) and 98.69% (active).
• Demonstrated high classification accuracy for lower limb dyskinesias: 99.45% (passive) and 99.63% (active).
• The model effectively differentiated dyskinesia severity across various Brunnstrom stages.

Conclusions:

• The combined NIRS and machine learning approach provides an accurate and automated method for assessing post-stroke dyskinesias.
• This technology shows significant potential for real-time monitoring of dyskinesias.
• The system can serve as a valuable tool for guiding personalized stroke rehabilitation training.