Brain Tumor Classification using MobileNet Convolutional Block Attention Module

Brain tumors are very deadly disease. The first step to investigating brain tumors is CT scan or MRI. The MRI images are then referenced for further investigation. It is very important to locate the tumors effectively. The manual methods of observing tumor location and shape can lead to wrong treatment. Automated methods such as Deep Learning helps to evaluate the type of the tumor and shape of the tumor effectively. In this research we used MobileNet as the pretrained network for the feature extraction process and convolutional block attention model for the visualization. Current study trained four models MobileNet CBAM, MobileNet SE, MobileNet ECA, MobileNet. The model was trained on Figshare dataset which include four classes, glioma, meningioma, no tumor and pituitary. We used 4,557 images to train the model and 1,311 images to test it. The proposed model achieved an accuracy of 95%. The CBAM attention-based visual explainer was used to show where the model was focusing its predictions, which made the model easier to understand. This method is a useful tool in clinical settings where quick and accurate diagnosis is important.