Integrating Machine Learning with Robotic Rehabilitation May Support Prediction of Recovery of the Upper Limb Motor Function in Stroke Survivors

Abstract

Motor impairment is a common issue in stroke patients, often affecting the upper limbs. To this standpoint, robotic neurorehabilitation has shown to be highly effective for motor function recovery. Notably, Machine learning (ML) may be a powerful technique able to identify the optimal kind and intensity of rehabilitation treatments to maximize the outcomes. This retrospective observational research aims to assess the efficacy of robotic devices in facilitating the functional rehabilitation of upper limbs in stroke patients through ML models. Specifically, clinical scales, such as the Fugl-Meyer Assessment (A-D) (FMA), the Frenchay Arm Test (FAT), and the Barthel Index (BI), were used to assess the patients’ condition before and after robotic therapy. The values of these scales were predicted based on the patients’ clinical and demographic data obtained before the treatment. The findings showed that ML models have high accuracy in predicting the FMA, FAT, and BI, with R-squared (R2) values of 0.79, 0.57, and 0.74, respectively. The findings of this study suggest that integrating ML into robotic therapy may have the capacity to establish a personalized and streamlined clinical practice, leading to significant improvements in patients’ quality of life and the long-term sustainability of the healthcare system.

Keywords

• stroke
• upper limbs
• robotic rehabilitation
• machine learning
• Fugl-Meyer Assessment (FMA)
• Frenchay Arm Test (FAT)