Explainable AI (XAI) based verification for brain tumor classification using deep learning techniques

Brain tumors pose a significant health concern, but their slow growth makes them difficult to detect. Doctors typically identify brain tumors by analysing the symptoms of a patient, medical history, and conducting tests. Diagnosis could be a lengthy process, without a way to confirm the doctor's recommendations. To address the issues related to time and verification, this paper proposes the incorporation of an appropriate deep learning model with an explainable artificial intelligence (XAI) approach. XAI is used to verify the outcomes of the proposed suitable deep learning models. Brain tumor image dataset downloaded from Kaggle is used for training and testing the model. 5618 images are used to train and 1405 images are used to test the model. Hence, total 7023 images are used. This paper identifies the best five deep learning models i.e. ResNet50, EfficientNetV2B0, ConvNeXtTiny, ResNet101, EfficientNetB1 based on Accuracy Score, Average Loss, Precision, Recall, and F1-Score. The Explainable AI based Grad-CAM method is applied on these best five models. It will be helpful to identify a single best model based on the basis of exact match of the actual and predicted classes. Results indicate that the best performing deep learning model is ResNet50, because its accuracy score is 96.08%. After that by applying the XAI Grad-CAM method, in the ResNet50 model it is verified that an exact match of the predicted class with the actual class at 88%, with a revision of classes 9%, and uncertain classes 3%, which are much less than the other competitive deep learning models.