Convolutional Redistribution Network for Multi-view Medical Image Diagnosis

Medical data such as Computed Tomography (CT), X-ray, and Magnetic Resonance Imaging (MRI) are integral elements of medical diagnosis. Deep learning has become common in computer-aided diagnosis, however, most of these models use only single-modal data as input and cannot take full advantage of data from different modalities for diagnosis. In addition, most of the existing multi-view models only fuse the results on a single view, without fully exploring the relationships between multi-view data. To better explore the correlation between data from different modalities, we propose a generic multi-view classification model for computer-aided diagnosis on multi-view medical images. With attention mechanism, the proposed model automatically extracts essential information from multi-view data to generate a series of "good and diverse" pseudo views for integration. The experiment results show that proposed model achieves good performance on pancreatic tumor classification task as well as the OrganMNIST3D classification task of the MedMNIST public datasets.