PrimSeq: A deep learning-based pipeline to quantitate rehabilitation training

Area of Science:

• Neurorehabilitation
• Biomedical Engineering
• Machine Learning

Background:

• Stroke rehabilitation aims to improve motor function through training, but optimal training doses remain unclear due to measurement challenges.
• Animal studies suggest high volumes of functional motion training significantly enhance upper extremity recovery post-stroke.
• Current methods lack practical tools to quantify functional motions during human rehabilitation, hindering dose-response studies.

Purpose of the Study:

• To introduce PrimSeq, a novel pipeline for classifying and quantifying functional motions in stroke rehabilitation.
• To enable accurate measurement of training doses for upper extremity recovery after stroke.
• To overcome limitations in current methods for assessing rehabilitation activity.

Main Methods:

• Integration of wearable sensors to capture upper-body kinematics.
• Application of a deep learning model for predicting motion sequences from sensor data.
• Development of an algorithm to classify and count elemental functional motions within rehabilitation activities.

Main Results:

• PrimSeq accurately decomposes complex rehabilitation activities into discrete functional motions.
• The pipeline demonstrates superior performance compared to existing machine learning approaches.
• Quantification of motions by PrimSeq is significantly faster and more cost-effective than manual expert analysis.
• Successful validation of PrimSeq on stroke patients with varying degrees of upper extremity impairment.

Conclusions:

• PrimSeq offers a practical and accurate solution for measuring functional motion doses in stroke rehabilitation.
• This methodology is crucial for advancing quantitative dosing trials to optimize upper extremity recovery.
• The tool has the potential to standardize and improve the efficacy of stroke rehabilitation training.