ABSTRACT Brain tumors (BT) are considered a major health challenge around the world, which needs early detection; therefore, effective treatment strategies can be planned. This kind of cancer greatly diminishes the patient's quality and lifespan, which opens a gateway for early diagnosis and effective treatment. The medical professionals need assistance with this difficult and error‐prone process; it is also mandatory to augment the interpretability and accuracy of the recognition model. To achieve such a goal, a hybrid deep learning model superior with explainable AI is introduced in the proposed framework, which performs brain tumor classification and model interpretation from MRI. The proposed study involves four key steps: Pre‐processing, segmentation, classification, and analysis. The input images are initially pre‐processed using a median‐boosted Kuan Filtering (Me‐KF) to remove any noise in the data and improve the subsequent segmentation procedure. After pre‐processing, the Extended Multi‐Inception Attention U‐Net (ExMIAU‐Net) technique is added to effectively separate the brain tumor region. Finally, a deep learning method based on Convolution Attentive assisted EfficientNetB0 (CA‐EfficientNetB0) is presented to categorize the many categories of brain tumors, comprising gliomas, meningiomas, pituitary tumors and normal tumors. This model uses Shapley additive explanation (SHAP), Local interpretable model‐agnostic explanations (LIME), and Gradient‐weighted Class Activation Mapping (Grad‐CAM) for model interpretation. The proposed model uses a brain tumor classification dataset. In the results section, the proposed model is compared to many other prevailing schemes and it achieves 99. 45% accuracy, 99. 16% precision, 98. 97% recall, and a 99. 06% F1‐score. The results show that an efficient, interpretable, robust and better model is developed for brain tumor classification.
Mazhar et al. (Fri,) studied this question.