RECOMMENDATION SYSTEM FOR PHYSICAL REHABILITATION SUPPORT BASED ON THERAPEUTIC IMAGE CORRECTION

Abstract

The paper presents the development and implementation of a recommendation system for supporting physical therapist decision-making in the rehabilitation of patients with upper limb mobility impairments. The recommendation system is integrated directly into the physiotherapist's web application built on React, which is a component of a comprehensive AR rehabilitation system with art therapy on the Magic Leap 2 headset. The system provides automatic collection of patient motor activity data and prediction of joint flexion angles based on interval mathematical models constructed from previous rehabilitation session data. Unlike traditional approaches, the proposed solution relies not on current measurements but on the analysis of predicted recovery dynamics, enabling proactive correction of the rehabilitation process.

The key mechanism for implementing recommendations is the correction of the therapeutic coloring image that the patient draws on a virtual AR canvas during art therapy sessions. The recommendation module analyzes predicted joint angle dynamics and, upon detecting a suboptimal recovery trajectory, generates recommendations for modifying the spatial distribution of image elements to stimulate targeted movements in the required anatomical plane. The system employs a multi-stage AI pipeline in which a textual recommendation is automatically transformed by a language model into an image generation prompt, after which a generative model creates the corresponding art content delivered to the AR canvas via the physiotherapist's web application. Experimental validation on two patients confirmed system operability: therapeutic image correction based on recommendations accelerated the shoulder joint flexion recovery rate by 1.9–3.9 times and reduced the predicted number of sessions to reach the normative range by 12–18%.