Brain Tumor Classification using Multi-Resolution Averaged Spatial Attention Features with CBAM and Convolutional Neural Networks

Medical imaging plays a vital role in diagnosis and understanding of different medical conditions, especially with regard to neurology. The automation of medical imaging analysis has been substantially improved over the last few years due to the integration of advanced technologies like Deep Learning. Developing a system for the classification of brain tumors is the primary objective. It involves training the deep learning model of a dataset from brain scans that are labeled with tumor types. Recently, pre-trained models have been applied where features are typically collected from the very bottom layer after the flatten layer that differs between natural and medical images. This work suggests a multi-level feature extraction using spatial features and channel averaging methods for early brain tumor identification in order to address this issue and achieve a classification accuracy of 96.70% for the test data