ADVANCING MULTI-MODAL BRAIN TUMOR CLASSIFICATION AND EXPANDING BEYOND TRADITIONAL CATEGORIES

Abstract

Recent developments in computer vision and machine learning have greatly enhanced medical image analysis, particularly in the field of identifying and categorizing brain tumors. This study explores various deep learning models' effectiveness for multi-modal brain tumor classification, considering a dataset comprising 4,478 MR images categorized into 44 distinct classes, including diverse brain tumor types and normal brain images. The research focuses on meticulous data preprocessing, leveraging pre-trained convolutional neural networks (CNNs), and employing transfer learning to fine-tune model architectures. Comprehensive experiments involving model training, validation, and testing reveal the effectiveness of models such as EfficientNetB5, ResNet50, VGG16, VGG19, MobileNetV3, DenseNet201, and InceptionV3. Evaluation criteria such as F1-score, accuracy, precision, and recall show how well the models perform across a range of brain tumor classes. The study highlights the superiority of the EfficientNetB5 model in achieving an accuracy of 95%, showcasing its potential as a cost-effective and reliable tool for assisting radiologists in brain tumor diagnosis. The results aim to promote early and accurate brain tumor diagnosis for better patient care by improving the efficiency and accuracy of deep learning models for brain tumor categorization.

Keywords – Tumor Classification, Deep Learning, Convolutional Neural Networks (CNNs), Medical Image Analysis, EfficientNetB5, Multi-modal Imaging, Transfer Learning.