AN INTELLIGIBLE AI-DRIVEN DECISION SUPPORT SYSTEM FOR POSTSTROKE MOBILITY ASSESSMENT.

Area of Science:

• Neurology
• Rehabilitation Medicine
• Artificial Intelligence

Background:

• Long-term mobility impairment is a common consequence for stroke survivors, necessitating extensive medical and physiotherapy.
• Accurate assessment of therapeutic success is crucial but challenging due to the complexity of mobility disorders and a growing demand for expert clinical services.
• Staff shortages and increasing patient numbers pose significant challenges in providing adequate post-stroke care and mobility assessment.

Purpose of the Study:

• To investigate the efficacy of machine learning algorithms in replicating expert-level mobility assessments using gait data from stroke patients.
• To develop an automated system that supports post-stroke mobility evaluation and provides interpretable feedback on assessment generation.
• To address the challenges posed by staff shortages and patient load in rehabilitation settings.

Main Methods:

• 100 hemiparetic stroke patients underwent clinical evaluations and instrumented gait analysis.
• An interdisciplinary expert board assigned a Stroke Mobility Score based on comprehensive gait data.
• Two regression models, a decision tree and a multilayer perceptron neural network, were trained on 680 extracted gait features.

Main Results:

• Both machine learning models demonstrated good to very good coefficients of determination in replicating expert mobility scores.
• Interpretable decision trees and neural network explanations identified key gait features crucial for mobility assessment.
• Automated assessments generated by the models showed strong agreement with expert evaluations.

Conclusions:

• Machine learning models can accurately reproduce expert mobility assessments from gait data in stroke patients.
• The developed system offers objective feedback and supports therapists in evaluating post-stroke mobility.
• Synergistic collaboration between AI systems and clinicians can enhance diagnostic quality and objectify therapeutic goals in stroke rehabilitation.