Exploring the State-of-the-Art Algorithms for Brain Tumor Classification Using MRI Data

This study presents a comprehensive approach for brain tumor classification using magnetic resonance imaging (MRI) and state-of-the-art image processing and deep learning techniques. MRI images provide high-resolution structural details of the brain, which are crucial for early detection and classification of brain tumors. The proposed methodology begins with image acquisition followed by a pre-processing pipeline involving grayscale conversion, noise reduction, normalization, and contrast enhancement. Data augmentation is performed to improve model generalization. K-means clustering is then applied for unsupervised segmentation to isolate tumor-affected regions. These segmented images are used to train a convolutional neural network (CNN) designed with multiple convolutional, batch normalization, ReLU, and pooling layers, culminating in a fully connected softmax classifier. The dataset is divided into training and validation subsets, and the model is trained using stochastic gradient descent with momentum (SGDM). Performance is evaluated using key metrics including accuracy, precision, recall, F1 score, Dice coefficient, and ROC. Results demonstrate high classification accuracy with a mean precision of 93.84%, F1 score of 93.68%, and mean Dice score of 93.68%. This pipeline shows potential in aiding medical professionals for accurate and automated brain tumor detection and classification, enhancing diagnostic efficiency and reducing subjectivity in clinical decisions.

Index Terms— Brain tumor, MRI, pre-processing, segmentation, CNN, classification, performance.