Design and Performance Evaluation of AI-Guided Therapeutic Machine for Wound Rehabilitation

Abstract

This study presents the design and evaluation of an AI-guided thermal therapy recommendation system for personalized wound rehabilitation. The system classifies wounds into seven types—ulcer, laceration, burn, incision, bruise, abrasion, and puncture—and identifies healing stage (fresh or healing) using a MobileNetV2-based deep learning model trained on 9,800 annotated wound images. Unlike existing AI wound classification systems that end at detection or grading, this approach directly maps classification results to specific hot or cold therapy parameters—temperature, airflow, duration, and frequency-based on NICE, WHO, and Journal of Wound Care guidelines. A “Special Recommendation” layer provides context-specific care instructions for each wound type–stage combination. The model achieved 95% training accuracy, 70% validation accuracy, and 93% therapy recommendation accuracy. A user-friendly web application enables real-time, automated, evidence-based prescriptions, offering potential benefits such as shorter healing times, reduced infection rates, and improved resource use, particularly in low-resource settings. While construction of the therapeutic machine is ongoing, detailed design parameters for adjustable temperature and ventilation are provided, demonstrating the feasibility of integrating AI classification with automated therapy delivery. These results highlight the system’s potential to improve clinical efficiency and patient outcomes, making it a scalable solution for modern wound care management.