DERMA DOCTOR Développement d'un système intelligent pour le diagnostic des maladies dermiques

Abstract

This thesis presents the development of an intelligent model for automatic detection of dermatological diseases, particularly skin lesions, using Artificial Intelligence (AI) and Deep Learning techniques. Faced with challenges related to manual diagnosis, including its slowness, high costs, and potential errors, we designed a system based on a MobileNetV2 architecture, which provides a balance between accuracy and speed. Using the HAM10000 dataset, which contains images of seven types of dermatological lesions, the model was trained in two phases: an initial training with frozen layers to extract essential features, followed by fine-tuning to optimize performance. This process resulted in a global accuracy of 82%, demonstrating the effectiveness of the proposed model. The model was then integrated into an interactive web application deployed via TensorFlow.js, allowing users to submit images of lesions directly through a browser to receive instant predictions. This work marks a significant step toward an automated and accessible solution for the early detection of dermatological diseases, with potential improvements through future clinical validation and the integration of advanced features.