Brain Tumor Diagnosing through Neural Networks

Brain tumor diagnosis from Magnetic Resonance Imaging (MRI) is a critical task, as precise detection of brain tumors helps improve clinical outcomes and patients’ quality of life. The objective of this project was to develop a deep neural network capable of classifying brain tumors from MRI scans with a target accuracy over 97%. For model training, we utilized the Brain Tumor MRI Dataset, which contains 7023 MRI images split across four classes: glioma tumor, meningioma tumor, pituitary tumor, and no tumor. We explored two variations of a custom convolutional neural network architecture, which leveraged depthwise separable convolution along with residual connections. Additionally, we fine-tuned six different models based on the pre-trained EfficientNet-B0 architecture. All models were evaluated using accuracy, precision, recall and f1-score metrics, and their performance was further assessed with the use of a confusion matrix. Our findings revealed that the best performing models, fine-tuned EfficientNet-B0 variants, achieved over 98% overall accuracy. These results demonstrate the effectiveness of fine-tuned deep learning models for high-accuracy medical image classification, surpassing our target performance and potentially contributing to faster and more reliable diagnostic tools in clinical settings.