Robust Methods for Real-Time Diabetic Foot Ulcer Detection and Localization on Mobile Devices

Area of Science:

• Medical imaging
• Artificial intelligence
• Diabetic foot ulcer detection

Background:

• Diabetic foot ulcers (DFU) require timely screening and localization for effective management.
• Current screening methods rely on manual detection by podiatrists.
• Existing automated solutions often lack real-time localization capabilities.

Purpose of the Study:

• To develop and evaluate deep learning models for real-time diabetic foot ulcer localization.
• To create a robust AI system for DFU detection and localization.

Main Methods:

• Collected an extensive dataset of 1775 DFU images.
• Utilized a Faster R-CNN with InceptionV2 model incorporating two-tier transfer learning.
• Performed five-fold cross-validation for model assessment.

Main Results:

• Achieved a mean average precision of 91.8% for DFU localization.
• Demonstrated a fast inference speed of 48 ms per image.
• Model size was optimized to 57.2 MB.

Conclusions:

• Deep learning effectively enables real-time localization of diabetic foot ulcers.
• The developed model shows robustness and practicality for real-time prediction on devices like NVIDIA Jetson TX2 and smartphone apps.
• Further improvements are anticipated with larger datasets.