Dense-ShuffleGCANet: An Attention-Driven Deep Learning Approach for Diabetic Foot Ulcer Classification Using Refined Spatio-Dimensional Features

This project presents Dense-ShuffleGCANet, an advanced deep learning-based diagnostic framework for Diabetic Foot Ulcer (DFU) classification, leveraging both spatial and channel attention mechanisms. We utilize four high-performance architectures—DenseNet-169, ShuffleNet, DenseNet+CCDGS (Channel-wise Cross-Domain Global Spatial), and DenseNet+CCDGS+TA (Triplet Attention)—trained on the publicly available Kaggle DFU dataset, achieving classification accuracies exceeding 95%. These models effectively distinguish between normal and abnormal foot conditions by capturing refined spatio-dimensional features. To ensure practical usability, we have developed a Flask-based web application integrated with HTML, CSS, and JavaScript. Users can register, log in, upload foot images, and receive real-time classification results, enabling remote screening and early intervention. This system serves as a scalable and accessible tool for healthcare providers and patients, offering an accurate and efficient solution for automated DFU detection, ultimately contributing to improved diabetic care and ulcer prevention.

KeywordsDiabetic Foot Ulcer (DFU), Deep Learning, DenseNet-169, ShuffleNet, CCDGS, Triplet Attention, Spatio-Dimensional Features, Flask Web App, Medical Image Classification, Abnormality Detection, Remote Diagnosis, Healthcare AI.