Lightweight CNN-Based Brain Tumor Classification from MRI

Abstract

Automated brain-tumor classification using Magnetic Resonance Imaging (MRI) can support image-based screening and clinical assessment. This paper presents a compact Convolutional Neural Network (CNN) for classifying two-dimensional brain MRI images into four categories: Glioma, Meningioma, Pituitary Tumor, and No Tumor. The model was trained and evaluated using a publicly available, multi-source collection of 7,200 images. The supplied test directory was retained as a held-out image-level test set, while the supplied training directory was divided into stratified training and validation subsets. The proposed model achieved an overall accuracy of 91.9%, with macro-averaged precision, recall, and F1-score values of approximately 91.9%, as well as a mean one-vs-rest AUC of 0.95. Based on the retained normalized timing records, the proposed CNN was used as the 1.00× throughput reference, compared with 0.60× for DenseNet169 and 0.65× for VGG19. These ratios indicate that the proposed CNN achieved 66.7% and 53.8% higher throughput than DenseNet169 and VGG19, respectively. The trained model was also integrated into an implemented Flutter–Flask–Firestore prototype that supports authenticated image uploads, server-side preprocessing and inference, prediction storage, and report retrieval. Functional testing verified these operations. However, end-to-end latency, concurrent-user capacity, patient-level generalization, external validation, and clinical utility remain to be established.