Radio Frequency based In-vivo Cognitive Load Detection Using Multivariate Features and Machine Learning

The detection of cognitive load (CL) has emerged as a significant research challenge in recent years. Most traditional techniques that rely heavily on physiological and audio-visual sensors are both privacy-invasive and computationally complex. The complexities of synchronization, data alignment, and accessibility limitations can potentially increase noise and error rates, thereby compromising the accuracy of CL estimates. This study presents a multi-modal, non-invasive, and privacy-preserving Radio Frequency (RF) sensing technique to overcome these limitations and enhance the reliability of CL estimation. The RF sensors are developed to capture blood flow changes in specific brain regions with high spatial resolution. The novel approach adopts RF sensing to estimate in-vivo CL variations utilizing pupillometry as a baseline. The in-vivo audio-only (AO) and audio-visual (AV) trials are conducted in controlled and uncontrolled environments with participants to comprehend target speech with varying background noise levels. Machine Learning (ML) and Deep Learning (DL) methodologies are evaluated for CL classification using RF statistical features with iterative features selection to build a robust feature set. The binary classification achieves an accuracy of 88% and 75% for AO and AV trials. The multi-class classification resulted in an accuracy of 81% and 69% for AO and AV trials. The proposed RF sensing can be utilized to assess the listening effort and CL of hearing-aid users. These RF measurements can be adapted to regulate real-time speech enhancement in hearing aids, tailored to the user's CL and complexity of the acoustic environment.