Automated Thermal Imaging for Diabetic Foot Risk Assessment

Diabetic foot is a prevalent complication affecting approximately 25% of individuals with diabetes, often leading to ulcers and amputations. Thermography is established as a non-invasive technique for early detection and monitoring of such complications by analyzing plantar temperature distribution and asymmetry. However, current software solutions are either dependent on specific camera systems or lack transparency and accessibility, limiting their clinical applicability. This study presents the development and validation of an automated tool for segmenting anatomical regions of interest (ROIs) in thermal foot images to support early diagnosis and monitoring of diabetic foot complications. The proposed system uses an object detection model trained on 92 annotated thermal images from two camera models. The algorithm identifies foot regions, applies edge detection and temperature-based thresholding, and extracts five ROIs per foot, using geometrical and temperature cues to define anatomical boundaries. The model was integrated as an API into the Vizbodx Thermal Analysis platform, enabling interactive visualization and refinement of the segmented ROIs. Evaluation using the Dice Similarity Coefficient showed high agreement with manual annotations across all ROIs (Dice > 0.75), with minimal segmentation failures. The tool also reduced the time and effort required for manual adjustments. The proposed solution offers a robust and efficient method for anatomical segmentation of thermal foot images. Its performance and integration into an existing platform demonstrate its potential as a practical support tool for diabetic foot screening and monitoring.