Design and Implementation of A Custom Convolutional Neural Network for Classifying Brain Magnetic Resonance Imaging Scans into Tumor Types

Brain tumours pose a critical healthcare challenge globally due to their potential for rapid progression and diagnostic complexity. In this research, we present a custom-built convolutional neural network (CNN) designed from scratch for the automatic detection and classification of brain tumours from magnetic resonance imaging (MRI). The model classifies images into four categories: glioma, meningioma, pituitary tumour, and no tumour. A total of 7024 MRI images were utilized, with a 90:10 train-test split. Performance was evaluated using metrics including accuracy, loss, precision, recall, and F1-score. Our model achieved a test accuracy of 96%, outperforming popular pretrained models including VGG16, ResNet50, and MobileNetV2. Notably, our CNN model uses smaller image dimensions (150×150) and does not rely on data augmentation, leading to reduced memory consumption. The study includes a comparative analysis and highlights the model's potential in supporting early and reliable diagnosis, particularly in resource-limited clinical settings.