Deep Learning in Medical Imaging: CNN-Based Approach for Multi-Class Brain Tumor Classification

Brain tumors rank as one of the most serious and potentially fatal conditions and the importance of early detection in addressing the ailment cannot be overlooked. Our research paper proposes a technology-enhanced approach to brain tumor classification using a Convolutional Neural Network (CNN) to ease the work of healthcare providers by providing them with effective tools for diagnosis. Based on MRI images obtained from the Kaggle database, this research aims to classify brain tumors correctly into four classes Glioma, Meningioma, Pituitary, and No Tumor. The proposed CNN has an architecture that consists of several convolutional layers for full-feature extraction and two fully connected layers for fine classification. The model employed categorical cross-entropy as the loss function and accuracy as the benchmark measure of performance. Training versus validation accuracy and loss graphs were also generated to assess the performance, stability, and generalization of the model. Our paper addresses the medical imaging processing in improving the diagnosis accuracy which is also a step forward towards the development of automated tumor detection systems. In the model that we have developed and implemented, we were able to achieve an accuracy of $\mathbf{9 6. 5 6 \%}$ and thus believe that it will be useful for automated medical diagnosis and can assist doctors in many different ways.