Apprentissage Profond pour identifier les cellules cancéreuses dans des images médicales

Abstract

Brain tumors are a complex and dangerous form of cancer, where early diagnosis is crucial to improving patients’ chances of survival. Traditionally, doctors have relied on clinical examinations and medical imaging techniques such as MRI. However, these methods, although effective, remain subjective and dependent on human expertise, which can lead to variations in diagnoses. The use of computer-aided diagnostic systems, notably via deep learning techniques, makes the process more objective, reproducible and accurate. These systems are based on convolutional neural networks (CNNs) capable of detecting complex patterns in medical images. In this work, we proposed a brain tumor classification system ,we exploit transfer learning, where the ResNet50 model was used as a base model reinforced by the Convolutional Block Attention Module (CBAM) . This module enables the model to focus on the most relevant regions of the images, thus improving tumor detection. We also applied data pretreatment techniques to improve model robustness.