Artificial intelligence exercise recommendations in knee osteoarthritis rehabilitation: ChatGPT-4o and Gemini Advanced example

Ömer Alperen Gürses; Anıl Özüdoğru; Figen Tuncay; Caner Karartı

doi:10.54005/geneltip.1634118

Artificial intelligence exercise recommendations in knee osteoarthritis rehabilitation: ChatGPT-4o and Gemini Advanced example

Authors : Ömer Alperen Gürses, Anıl Özüdoğru, Figen Tuncay, Caner Karartı

Pages : 487-492

Doi:10.54005/geneltip.1634118

View : 71 | Download : 45

Publication Date : 2025-06-30

Article Type : Research Paper

Abstract :Abstract Aim: This study aimed to comparatively evaluate the propensity of the large language models ChatGPT-4o and Gemini Advanced to recommend personalised exercise based on patients\\\' assessment data in knee osteoarthritis rehabilitation. Methods: This observational study included 40 patients diagnosed with knee OA according to the American College of Rheumatology criteria. Demographic data, pain levels, range of motion, muscle strength, functional status, and balance were assessed using standardized clinical tests. ChatGPT-4o and Gemini Advanced generated three-phase rehabilitation programs based on these assessments. Exercise recommendations were analyzed across 12 parameters, and statistical comparisons were conducted using the Mann-Whitney U test and Spearman’s correlation (p<0.05). Results: ChatGPT-4o demonstrated statistically significant differences in 7 parameters: Phase 1 (quadriceps muscle strength, knee flexion angle, knee extension angle, and four-square step test; p=0.017, p=0.012, p=0.033, p=0.043), Phase 2 (quadriceps muscle strength and Lysholm scale; p=0.032, p=0.040), and Phase 3 (quadriceps muscle strength; p=0.007). In contrast, Gemini Advanced exhibited significant differences in only 2 parameters: Phase 1 (Lysholm scale score; p=0.044) and Phase 3 (quadriceps strengthening exercise; p=0.047). ChatGPT-4o appeared to integrate patient assessment data more effectively, but both models showed limitations in personalization. Conclusions: While ChatGPT-4o and Gemini Advanced show potential for designing personalized knee OA rehabilitation programs, their recommendations remain constrained. Further improvements in dataset quality, real-time medical knowledge integration, and domain-specific training are needed to enhance their clinical utility.
Keywords : Yapay zeka, ChatGPT, Gemini, büyük dil modelleri, fizyoterapi, diz osteoartriti, rehabilitasyon programı

ORIGINAL ARTICLE URL