Development of a Low-Cost Embedded Thermal Imaging System with Artificial Intelligence for Diabetic Foot Assessment

Description

DiaFoot AI is a portable embedded thermal imaging system developed to support diabetic foot assessment through standardized plantar image acquisition and artificial intelligence. The prototype integrates an MLX90640 thermal sensor, an NVIDIA Jetson embedded platform, a custom 3D-printed foot-positioning structure, and locally executed AI-based image classification. The mechanical support maintains the plantar surface in a stable orientation while the thermal camera is positioned approximately 40–45 cm from the sole, reducing variability related to patient movement, imaging angle, and camera distance. The system operates offline and enables patient registration, real-time thermal image acquisition or image upload, AI-based classification, visualization of prediction confidence, and automatic PDF report generation. Functional validation using 1,000 independent images achieved an overall classification accuracy of 95.0%, while stratified 10-fold cross-validation yielded a mean validation accuracy of 95.03%. Average inference time was approximately 1–2 s per image. Finite-element analysis showed limited structural deformation, with a maximum displacement of 0.5848 mm. DiaFoot AI is intended as a clinical decision-support and screening platform for hospitals, primary healthcare institutions, and resource-limited settings.

Institutions

• Corporación Universitaria del Huila

  Huila Department, Neiva
• Universidad Surcolombiana

  Huila Department, Neiva
• Universidad Incca de Colombia

  Bogota D.C., Bogotá
• Fundación Escuela Tecnológica de Neiva

  Huila Department, Rivera

Categories

Licence