Plantar Thermogram Analysis Using Deep Learning for Diabetic Foot Risk Classification

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Diabetology

Background:

• Diabetic foot ulcers are a significant complication of diabetes, leading to severe morbidity.
• Early identification of patients at risk is crucial for preventing ulceration and amputation.
• Thermography offers a non-invasive method for assessing foot temperature variations, indicative of underlying pathology.

Purpose of the Study:

• To investigate the efficacy of combining thermography with deep learning for risk stratification of diabetic patients prone to foot ulcers.
• To develop and validate a deep learning model for analyzing plantar thermograms to identify at-risk individuals.

Main Methods:

• Prospective data collection of clinical information and plantar thermograms from adult diabetic patients undergoing screening.
• Analysis of 153 thermal images using a deep learning algorithm, with a neural network trained on a balanced dataset (80% training, 20% validation).
• Validation of the trained model on a separate testing dataset (55 images) with a focus on maximizing sensitivity for detecting at-risk feet.

Main Results:

• The study included 153 patients with a mean age of 63.1 years; 52.3% were female.
• The deep learning model, using five-fold cross-validation, achieved an overall accuracy of 71.8%, sensitivity of 81.2%, and specificity of 64.0%.
• The Matthews correlation coefficient was 0.46, indicating moderate agreement between predicted and actual risk categories.

Conclusions:

• Thermography combined with deep learning demonstrates potential as a screening tool for diabetic foot ulcer risk stratification.
• The developed model shows promise in identifying patients who require closer monitoring and intervention to prevent ulcer development.
• Further development and validation are warranted to optimize this approach for clinical practice.