Résumé : Brain tumors are masses or abnormal growths of cells within the brain or the central spinal canal with symptoms such as headaches, seizures, weakness or numbness in the arms or legs, changes in personality or behaviour, nausea, vomiting, vision or hearing problems and dizziness. Conventional diagnosis of brain tumour involves some tests and procedure which may include the consideration of medical history, physical examination, imaging tests (such as CT or MRI scans), and biopsy (removal and examination of a small piece of the tumor tissue). These procedures, while effective, are mentally strenuous and time demanding due to the manual examination of the brain scans and the thorough evaluation of test results. It has been established in lots of medical research that brain tumours diagnosed and treated early generally tends to have a better prognosis. Deep learning techniques have evolved over the years and have demonstrated impressive and faster outcomes in the classification of brain tumours in medical imaging, with very little to no human interference. This study proposes a model for the early detection of brain tumours using a combination of convolutional neural networks (CNNs) and extreme gradient boosting (XGBoost). The proposed model, named C-XGBoost has a lower model complexity compared to purely CNNs, making it easier to train and less prone to overfitting. It is also better able to handle imbalanced and unstructured data, which are common issues in real-world medical image classification tasks. To evaluate the effectiveness of the proposed model, we employed a dataset of brain MRI images with and without tumours.