A deep learning network method based on triple squeeze-and-excitation multi-scale attention for brain tumor detection

In clinical practice, the diagnosis of brain tumors is very much dependent on the experience of experienced doctors. Therefore, it is crucial to use computers to assist doctors in diagnosing and treating brain tumors. In this paper, we propose a brain tumor classification algorithm with the help of a convolutional neural network to save the doctor's diagnostic time and assist the doctor in quickly and accurately diagnosing three common brain tumors and normal brain images from MR images. The method is divided into image preprocessing and our proposed convolutional neural network. Firstly, the input image is subjected to a series of geometric transformations for image enhancement, which are used to improve the robustness of the neural network. Subsequently, the image passes through our lightweight neural network design, consisting of three convolution layers and a triple SEM attention mechanism for extracting the image feature information. Finally, we use a CAM heat map to analyze the network's decision-making basis for disease judgment. In the experimental results, our proposed diagnosis achieved a diagnostic accuracy of 95%; this experimental result proves the effectiveness of this method for computer-aided brain tumor detection.