Brain tumor detection from MRI images with explainable artificial intelligence methods

Within the scope of this study, it was aimed to detect brain tumors from MRI images with explainable artificial intelligence methods. It is thought that this study is important in terms of contributing to medical processes and providing a new literature on decision support systems. Within the scope of the study, GradCAM, LIME and Shapley visualization methods from CNN models were used. In the model created in the study, the classification was examined under 4 groups (No Tumor, Glioma, Meningioma, Pituitary). As a result of the study, GradCAM was effective in determining the general focus areas of the model; LIME provided a detailed explanation of the model's decisions; Shapley revealed the general performance and shortcomings of the model. Using these techniques together allows more data to be provided or necessary improvements to be made to ensure that the model works more reliably and effectively