Enhanced Automated Brain Tumour Detection Using Convolutional Neural Networks

Precise identification of brain tumors is important for treatment and outcomes. Manual analysis of MRI scans is very time-consuming and prone to errors. This paper explores the application of CNNs for automatic classification of brain MRI scans to help radiologists provide rapid and accurate diagnoses. The dataset was pre-processed to ensure consistency and generalizability through resizing, normalization, and data augmentation. Data augmentation methods include rotating, flipping, and brightening have been applied to deal with few numbers of training images, enhancing the model. Convolutional, pooling, and dropout layers in the CNN architecture are focused to achieve feature extraction, avoid overfitting, and recognize brain tumors. It has also been optimized using Adam with its testing process concerning accuracy, precision, and recall. Excellent generalization of $\mathbf{9 4 \%}$ training and $\mathbf{9 1 \%}$ validation. Good scores for the cases producing high precision and high recall validate the correctness in no false positives as well as false negatives of tumor detection by this model. Results improve efficiency and accuracy of the diagnostic process, aiding timeand consistency-strained radiologists. Larger sets and architecture refinement as future works could make it more relevantly clinically for testing across various settings.