Optimizing MEG-EEG Mapping in Resource-Constrained Non-Intrusive Bio-Magnetic Sensing Systems: A Data-Driven Approach

While Magnetoencephalography (MEG) and electroencephalography (EEG) are well-known neuroimaging techniques to capture a myriad of brain activities and stimulations, accessing conventional M/EEG devices is challenging. This is because of the bulky nature of the MEG machines, the need for magnetic shielding and cooling system, intrusive EEG electrodes, and various other complications involving preparing these devices to guarantee a clinical-grade signal acquisition. To address these issues, in this paper, we consider bio-magnetic sensing with emerging Magnetic Tunnel Junction (MTJ) sensors operating at room temperature that can map the sensed MEG to EEG signals, which can be helpful in this domain. However, such ultra-sensitive sensors are resource-constrained, and incorporating such MEG-EEG mapping needs to be optimized to balance the accuracy and computational/energy trade-off. Therefore, we adopt a data-centric approach to address this optimization problem. Furthermore, we conduct rigorous comparative analytics on prominent machine/deep learning models on a publicly available dataset to establish a baseline proof-of-concept that can be seamlessly integrated with the considered bio-magnetic sensing systems. Our research unlocks the possibility for real-time monitoring of brain activities and abnormality detection away from the clinical environment and complex hospital settings. Moreover, the compact size and low-power requirements for the considered MTJ sensor make it compatible with IoT and wearable devices.