Modified InceptionV3 Model for Brain Tumor Classification with Grad-CAM Explainability

The unrestrained growth of an abnormal cell cluster is the distinguishing characteristic of a brain tumor, which poses a highly dangerous situation. Inaccurate diagnoses of brain tumors will make it more challenging for them to receive treatment on time and reduce a patient's chances of life. Accurate tumor diagnosis is crucial for improving patient survival rates in medical imaging. Accordingly, this paper offers a modified pre-trained convolutional neural network (CNN) model for accurate brain tumor classification. For this purpose, four distinct pre-trained CNN models: VGG16, InceptionV3, DenseNet121, and MobileNetV2 are utilized for primary classification. The top classification layers of these models are removed and replaced with additional layers for the modification task. This procedure leads to the generalization of the modified pre-trained CNN models. In addition, we have constructed a customized CNN model including softmax classification. Following the analysis, it is found that the suggested model, which is a modified version of InceptionV3,achieves an impressive accuracy of 99.16%. Other models also yielded acceptable results: the modified MobileNetV2, VGG16, DenseNet121, and the customized CNN achieved accuracy of 98.16%,92.52%,99.09%, and 95.19%, respectively. To improve the interpretability of the proposed model's output, we employed Grad-CAM to visually represent the regions of the images the proposed model focuses on.