A CNN-based Approach for Multi-Classification of Brain Tumors

Abstract

Early diagnosis of brain tumor plays an important factor in extending the life expectancy of a patient. Therefore, an accurate and timely diagnosis of the type of brain tumor will allow adequate treatment planning and medical assistance. Radiologists commonly use magnetic resonance imaging (MRI) scans to detect and classify brain tumors. The current methods used in the medical field for diagnosis are time-consuming and prone to human error. In recent years, researchers have developed automated techniques for the segmentation and classification of MRI images resulting in a faster diagnosis process. Recent advancements in deep learning have shown greater efficiency in image recognition and classification tasks. In this paper, a convolutional neural network (CNN) (a widely used deep learning architecture for image classification tasks) is developed to classify MRI images into four brain tumor categories. Data augmentation is applied to the training dataset to generalize the images and avoid overfitting problem. Additionally, this paper compares the performance of various pre-trained models such as Vision Transformer (VIT), VGG19, ResNet50, Inception V3, and AlexNet50 with that of the proposed model. Each experiment then explores transfer learning techniques like fine-tuning and freezing layers. In the study, the proposed model yields the most efficient results with a classification accuracy of 94.72%.