Deep Learning-Based Model with XAI for Brain Tumor Classification and Segmentation Using MRI Images

Brain tumor classification & segmentation are crucial challenges in medical image analysis. Brain tumors are often incurable malignancies that arise from glial support cells. Medical study reveals that manual classification with human-assisted support might result in inaccurate prognosis and diagnosis because of the diversity and similarity of malignancies & normal tissues. Brain tumors are classified & segmented using T1, T1c, T2, and FLAIR MRI modalities. In this paper, we provide a technique for extracting brain tumours from 2D MRI images using several deep-learning tumor classification models (ResNet50, InceptionV3, VGG19, VGG16, DenseNet121, and EfficientNet). To give interpretability and explainability for the classification findings, the Lime (Local interpretable Model-Agnostic explanations) approach is also applied. During the segmentation step, we explore three widely used architectures: U-Net, SegNet, & ResUNet.The application of sophisticated classification models, XAI, and a fresh dataset establishes the framework for better tumor diagnosis and segmentation from 2D MRI images. In our experimental study, the ResNet50 model beat the other deep learning models, obtaining 96% accuracy. Furthermore, the suggested segmentation models ResUnet, Segnet, and U-Net have dice scores of 0.851, 0.772, and 0.7424, respectively.