Explainable artificial intelligence model to predict brain states from fNIRS signals.

Objective: Most Deep Learning (DL) methods for the classification of functional Near-Infrared Spectroscopy (fNIRS) signals do so without explaining which features contribute to the classification of a task or imagery. An explainable artificial intelligence (xAI) system that can decompose the Deep Learning mode's output onto the input variables for fNIRS signals is described here. Approach: We propose an xAI-fNIRS system that consists of a classification module and an explanation module. The classification module consists of two separately trained sliding window-based classifiers, namely, (i) 1-D Convolutional Neural Network (CNN); and (ii) Long Short-Term Memory (LSTM). The explanation module uses SHAP (SHapley Additive exPlanations) to explain the CNN model's output in terms of the model's input. Main results: We observed that the classification module was able to classify two types of datasets: (a) Motor task (MT), acquired from three subjects; and (b) Motor imagery (MI), acquired from 29 subjects, with an accuracy of over 96% for both CNN and LSTM models. The explanation module was able to identify the channels contributing the most to the classification of MI or MT and therefore identify the channel locations and whether they correspond to oxy- or deoxy-hemoglobin levels in those locations. Significance: The xAI-fNIRS system can distinguish between the brain states related to overt and covert motor imagery from fNIRS signals with high classification accuracy and is able to explain the signal features that discriminate between the brain states of interest.

Self-regulation of language areas using real-time functional MRI in stroke patients with expressive aphasia.

The objectives of this study were to test (i) If stroke patients with expressive Aphasia could learn to up-regulate the Blood Oxygenation Level Dependent (BOLD) signal in language areas of the brain, namely Inferior Frontal Gyrus (Broca's area) and Superior Temporal Gyrus (Wernicke's area), with real-time fMRI based neurofeedback of the BOLD activation and functional connectivity between the language areas; and (ii) acquired up-regulation could lead to an improvement in expression of language. The study was performed on three groups: Group 1 (n = 4) of Test patients and group 2 (n = 4) of healthy volunteers underwent the neurofeedback training, whereas group 3 (n = 4) of Control patients underwent treatment as usual. Language performance and recovery were assessed using western aphasia battery and picture naming tasks, before and after the neurofeedback training. Results show that the Test group had significant increase in activation of the Broca's area and its right homologue, while the Normal group achieved the greatest activation during neurofeedback. For the Test group both perilesional and contralateral activations were observed. The improvement in language ability of the test patients was not significantly greater than that of the control patients. Neurofeedback training in Aphasia patients induced significant activation of the Broca's area, Wernicke's area and their right homologues, although healthy individuals achieved greater activations in these regions than the patient groups. Training also activated perilesional areas of Rolandic operculum, precentral gyrus and postcentral gyrus for the Test patients significantly. However, lack of behavioral and symptom modifications in the Test group calls for improvements in the efficacy of the approach.

Functional Connectivity of Language Regions of Stroke Patients with Expressive Aphasia During Real-Time Functional Magnetic Resonance Imaging Based Neurofeedback.

Stroke lesions in the language centers of the brain impair the language areas and their connectivity. This article describes the dynamics of functional connectivity (FC) of language areas (FCL) during real-time functional magnetic resonance imaging (RT-fMRI)-based neurofeedback training for poststroke patients with expressive aphasia. The hypothesis is that FCL increases during the upregulation of language areas during neurofeedback training and that the training improves FCL with an increasing number of sessions and restores it toward normalcy. Four test and four control patients with expressive aphasia were recruited for the study along with four healthy volunteers termed as the normal group. The test and normal groups were administered four neurofeedback training sessions in between two test sessions, whereas the control group underwent only the two test sessions. The training session requires the subject to exercise language activity covertly so that it upregulates the feedback signal obtained from the Broca's area (in left inferior frontal gyrus) and amplifies the feedback when it is correlated with the Wernicke's area (in left superior temporal gyrus) using RT-fMRI. FC was measured by Pearson's correlation coefficient. The results indicate that the FC of the test group was weaker in the left hemisphere than that of the normal group, and post-training the connections have strengthened (correlation coefficient increases) in the left hemisphere when compared with the control group. The connections of language areas strengthened in both hemispheres during neurofeedback-based upregulation, and multiple training sessions strengthened new pathways and restored left hemispheric connections toward normalcy.

Brain-computer interfaces for neurorehabilitation.

Brain-computer interfaces (BCIs) enable control of computers and other assistive devices, such as neuro-prostheses, which are used for communication, movement restoration, neuro-modulation, and muscle stimulation, by using only signals measured directly from the brain. A BCI creates a new output channel for the brain to a computer or a device. This requires retrieval of signals of interest from the brain, and its use for neuro-rehabilitation by means of interfacing the signals to a computerized device. Brain signals such as action potentials from single neurons or nerve fibers, extracellular local field potentials (LFPs), electrocorticograms, electroencephalogram and its components such as the event-related brain potentials, real-time functional magnetic resonance imaging, near-infrared spectroscopy, and magneto-encephalogram have been used. BCIs are envisaged to be useful for communication, control and self-regulation of brain function. BCIs employ neurofeedback to enable operant conditioning to allow the user to learn using it. Paralytic conditions arising from stroke or other diseases are being targeted for BCI application. Neurofeedback strategies ranging from sensory feedback to direct brain stimulation are being employed. Existing BCIs are limited in their throughput in terms of letters per minute or commands per minute, and need extensive training to use the BCI. Further, they can cause rapid fatigue due to use and have limited adaptability to changes in the patient's brain state. The challenge before BCI technology for neuro-rehabilitation today is to enable effective clinical use of BCIs with minimal effort to set up and operate.