A Synergic Neural Network For Medical Image Classification Based On Attention Mechanism

For current research in using the deep learning method analyzing and processing medical image in CT or MRI format, identifying the category is challenging because the pathology of the medical images can only be distinguished by tiny difference since the medical image of human organ or tissue is highly homogeneous. The purpose of this paper is to design a convolutional neural network model to extract medical image features and establish a model to classify the medical images. We designed a synergic network based on attention mechanism, capturing the intra-class differences and similarities between classes. The whole pipeline consists of three parts: the input layer, attention based convolutional block and synergic network. By combining these three components, we trained an end-to-end model to classify the medical image. Moreover, through a carefully designed attention model, the network can adaptively pay attention to the region of interest of medical images, avoiding using expensive annotations like bounding boxes or part information labeling. We compare our model on the ADNI dataset(part of) and MRNet dataset with the current SOTA method; the results show our method achieves SOTA performance.