Multi-Class Brain Tumor Detection from MRI Scans Using CNN and Transfer Learning Techniques Analysis

Brain tumors represent one of the most urgent challenges in medical diagnosis, requiring timely and accurate identification to improve patient outcomes. Magnetic Resonance Imaging (MRI) remains the preferred imaging modality due to its high resolution and non-invasive nature. While MRI offers significant diagnostic advantages, manual interpretation is prone to inconsistencies and may delay early detection, motivating the development of automated classification systems.

This paper presents a deep learning-based approach for multi-class brain tumor classification from MRI images. A custom Convolutional Neural Network (CNN) was developed as a baseline and compared against three established transfer learning architectures — ResNet18, DenseNet121, and EfficientNetB0 — on a balanced dataset comprising four diagnostic categories: glioma, meningioma, pituitary tumor, and no tumor. To ensure result reliability, models were evaluated using stratified k-fold cross-validation (k = 5) in addition to a held-out test set.

EfficientNetB0 achieved the highest mean classification accuracy of 99.1% on the benchmark dataset, while maintaining computational efficiency. These results demonstrate the potential of transfer learning as a proof-of-concept diagnostic aid; however, external validation on multi-institutional, diverse MRI datasets is required before any clinical deployment can be considered. It is important to note that these results were obtained on a well-curated single-source benchmark dataset and may not reflect performance on real-world heterogeneous clinical data.