Validation of a Trained AI Imaging Model for Detecting Diabetic Foot Deformities.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Diabetology

Background:

• Diabetes mellitus is a major cause of morbidity and mortality, with diabetic foot complications posing significant risks.
• Early detection of diabetic foot deformities is crucial to prevent ulceration, infection, and amputation.

Purpose of the Study:

• To develop and validate an AI-powered image analysis tool for early detection of diabetic foot structural deformities.
• To identify early visual risk markers such as callus, hallux valgus, and hammer toes.

Main Methods:

• Utilized YOLOv5 for object detection, a convolutional autoencoder for anomaly detection, and DenseNet201 for classification.
• Trained and validated the AI model on datasets of healthy and diabetic foot images.
• Integrated the AI model into the 'Foot at Risk' (FAR) mobile application.

Main Results:

• Achieved high accuracy in internal validation (91.1% anomaly detection, 88.57% classification).
• External validation showed 85.2% overall accuracy and 66.7% anomaly classification accuracy.
• Final evaluation demonstrated 88.57% accuracy, 90.47% precision, and an 86.11% F1 score.

Conclusions:

• The AI-driven solution shows promise for scalable, early diabetic foot deformity screening.
• Integration into mobile applications and community care models can facilitate timely intervention.
• Further development with larger datasets can enhance its utility in improving patient outcomes.