Abstract
There have been several attempts to use deep learning based on brain fMRIsignals to classify cognitive impairment diseases. However, deep learning is ahidden black box model that makes it difficult to interpret the process ofclassification. To address this issue, we propose a novel analytical frameworkthat interprets the classification result from deep learning processes. Wefirst derive the region of interest (ROI) functional connectivity network (FCN)by embedding functions based on their similar signal patterns. Then, using theself-attention equipped deep learning model, we classify diseases based ontheir FCN. Finally, in order to interpret the classification results, we employa latent space item-response interaction network model to identify thesignificant functions that exhibit distinct connectivity patterns when comparedto other diseases. The application of this proposed framework to the four typesof cognitive impairment shows that our approach is valid for determining thesignificant ROI functions.