The use of AI-powered thermography to detect diabetic foot ulcers: a validation study (Preprint)

Abstract

BACKGROUND Diabetic foot problems are among the most debilitating complications of diabetes mellitus. The prevalence of diabetes mellitus and its complications, notably diabetic foot ulcers (DFUs), continues to rise, challenging healthcare despite advancements in medicine. Traditional detection methods for DFUs face scalability issues due to inefficiencies in time and practical application, leading to high recurrence and amputation rates alongside substantial healthcare costs. Human Medical Thermography presents a viable solution, offering an inexpensive, portable method without ionizing radiation, which could significantly enhance disease monitoring and detection, including DFUs. OBJECTIVE The purpose of this study is to evaluate the efficacy of AI-powered thermography in detecting plantar thermal patterns that differentiate between adult diabetic patients without visible foot ulcers and healthy individuals without diabetes. METHODS This prospective cohort validation study recruited a random sample of 200 patients; 100 patients were healthy, and the other 100 were diagnosed with diabetes but without a visible foot ulcer. Participants completed a baseline study questionnaire to gather initial data. Following this, a Research Assistant prepared participants for thermal imaging, which was conducted to capture plantar thermal patterns. All collected data, including thermal images and questionnaire responses, were stored on a password-protected computer to ensure confidentiality and data integrity. RESULTS Participants were categorized into two groups: a healthy control group (n = 98) with no prior diabetes or PAD diagnosis and normal circulatory findings, and a diabetic group (n = 98) comprising patients with diabetes, regardless of peripheral circulatory status. Analysis of both feet revealed significantly greater differences between feet in the diabetic group compared to controls (control 0.47 °C ± 0.43°C vs diabetic 1.78 °C ± 1.58 °C, p < 0.001, CI 0.99, 1.63). These results identified clinically relevant abnormalities in 10% of the diabetic cohort, whereas no such findings were observed in the control group. We used a linear regression model to indicate that being diagnosed with diabetes is a significant predictor of abnormal temperature, while age and gender were not found to be significant predictors in this model. CONCLUSIONS DFUs pose a significant health risk for diabetes patients, making early detection crucial. This study highlights the potential of an AI-powered computer vision system in identifying early signs of diabetic foot complications by differentiating thermal patterns between diabetic patients without visible ulcers and healthy individuals. The findings suggest that the technology could improve early diagnosis and outcomes in diabetic foot care. Further research with larger and more diverse populations is essential to validate its effectiveness and applicability.