A New Approach for Analyzing Functional Neuroimaging Data Using a Combination of CNN-LSTM and Occlusion Sensitivity Analysis to Identify Important Brain Regions in Visual Mental Imagery

Aphantasia is a condition that inhibits the ability to create mental images. In recent years, researchers have discovered a potential correlation between aphantasia and other mental health conditions, such as autism. However, due to the intricacy of the cognitive task, there is a debate among scientists about the specific regions of the brain that are causally involved in our ability to create mental images. The objective of this study is to present a groundbreaking deep learning framework for spatio-temporal analysis of block-designed fMRI signals. This framework, which we have named the classification and explanation deep learning framework for fMRI Signals (CEDLF-fMRI), has been designed to meet the complex needs of fMRI signal analysis. Our testing has demonstrated that CEDLF-fMRI outperforms traditional methods in classifying fMRI signals from complex experimental conditions. Furthermore, our framework generates a 3D image that provides an in-depth explanation of its decision-making process. Utilizing the generic object decoding dataset, the proposed CEDLF-fMRI demonstrated exceptional performance by classifying previously unseen samples at an impressive rate of 99% for two distinct categories: perception and imagery. Additionally, CEDLF-fMRI suggests that Right Lingual Gyrus (RLG) plays a key role in its ability to tell whether a signal belongs to imagery class or not.