Deep Learning Framework for Early Detection, Risk Calculation, Localization and Analysis of Diabetic Foot Ulcers

Diabetic Foot Ulcers (DFUs) are among the most severe complications of diabetes, often leading to infection,

hospitalization, and amputation if not detected early. In this work, we present a hybrid diagnostic system that integrates deep

learning-based image classification with a clinically inspired rule-based risk assessment model. The proposed system employs an

EfficientNetB3 convolutional neural network to classify wound images into abnormal (ulcer) and normal categories. To improve

interpretability, Gradientweighted Class Activation Mapping (GradCAM) is used to highlight regions of interest in the input

image. In addition, a clinical rule engine evaluates patientspecific parameters such as Ankle-Brachial Index (ABI), blood oxygen

levels (SpO2), blood glucose, age, and diabetes duration to estimate ischemia risk and assign a severity stage. The system is

deployed using a Flask-based API with a responsive frontend interface and containerized using Docker for scalability.

Experimental results demonstrate effective classification performance and enhanced clinical relevance through risk

stratification. This hybrid approach provides both predictive accuracy and explainability, making it suitable for real-world

healthcare applications