Explainable artificial intelligence for functional brain development analysis: methods and applications

In the last decades, non-invasive and portable neuroimaging techniques, such as functional Near- Infrared Spectroscopy (fNIRS), have allowed researchers to study the mechanisms underlying the functional development of the human brain, thus furthering the potential of Developmental Cognitive Neuroscience (DCN). However, the traditional methods used for the analysis of infant fNIRS data are still quite limited. Here, I introduce new Fuzzy Cognitive Maps, called EFCMs, for Effective Connectivity (EC) analysis of infants’ fNIRS data. EFCMs can outline the interconnections between the cortical areas as well as specify the direction of EC. In contrast, to shed light on the activation level of the cortical regions, I developed a Multivariate Pattern Analysis (MVPA). The proposed MVPA is powered by eXplainable Artificial Intelligence (XAI), named eXplainable MVPA (xMVPA). The xMVPA is exemplified in a DCN study that investigates visual and auditory processing in six- month-old infants with a classification accuracy of 67.69 %. The xMVPA can identify patterns of cortical interactions formed in response to presented stimuli as hypothesised by the DCN frameworks. However, xMVPA can only analyse cross-sectional DCN studies, i.e. it is not able to analyse the temporal dynamics associated with a longitudinal DCN study. To this end, I developed a novel time-dependent XAI (TXAI) system based on Temporal Type-2 Fuzzy Sets (TT2FS). The TXAI system is exemplified on an empirical study using a real-life intelligent environments dataset to solve a time-dependent classification problem and attained a classification accuracy of 94.08%. The proposed TXAI system has the potential to inform the evolution of a process (such as functional brain development) using temporal trajectories which in turn may assist in the delineation of brain developmental trajectories.