Image Classification of Wounds in Diabetes Mellitus Patients Using Convolutional Neural Network (CNN) Algorithm

Abstract

Diabetes Mellitus (DM) is a chronic disease that can lead to serious complications such as chronic wounds that are difficult to heal and may result in amputation. Early detection of wound types is essential to prevent worsening conditions. This study aims to develop a wound image classification system for diabetic patients using the Convolutional Neural Network (CNN) algorithm with the MobileNetV2 architecture. The dataset consists of seven classes: diabetic wounds (gangrene, blister, pressure), common wounds (abrasion, contusion, burn), and normal skin. The data were obtained from secondary sources (Roboflow) and primary sources from Rumah Sakit Umum Madani Medan. The research stages include data acquisition, augmentation, preprocessing, model training, performance evaluation, and implementation into a web-based application. Evaluation results show that the model achieved a training accuracy of 98% and a validation accuracy of 81%, along with high precision, recall, and F1-score values across most classes. The system is implemented as a web application to assist in early wound type detection and provide initial treatment recommendations, thereby supporting efforts to prevent complications in diabetic patients.