AI-driven Integration of EEG and Motor Assessment to Explore Motor-Cognitive Interaction in Multiple Sclerosis

Fall risk assessment in people with multiple sclerosis (PwMS) is typically based on a series of clinical scales and questionnaires. In the past decade, instrumental analysis using wearable sensors has gained attention, although technical limitations, such as recording electroencephalography (EEG) while walking, still limit their applicability. We propose a novel framework for assessing cognitive-motor interaction in PwMS within a cross-sectional study, integrating wearable 8-channel EEG with an inertial measurement unit (IMU) sensor placed on participants' forearm. The study involved 13 PwMS (6 females; mean age: 59.3 ± 11.5 years; Kurtzke Expanded Disability Status Scale (median [25th-75th percentiles]: 5.5 [4.25-6.00]), who performed i) the timed up and go (TUG) in isolation (single-task motor, STm); ii)the TUG while performing a cognitive task of serial subtraction of 7s (motor-cognitive dual-task [DT] condition); iii) the serial subtraction for 90 while sitting. Motion data were used for automatic TUG segmentation via a hybrid convolutional neural network and long-short-term-memory model, which also allows the alignment and extraction of EEG epochs corresponding to each trial. EEG workload was quantified using the task load index (TLI), that quantifies cognitive workload defined as the ratio of the power spectral density (PSD) in theta (4-8 Hz) and alpha (8-12 Hz) bands. The model allowed accurate TUG segmentation, with a root-mean-squared-error (RMSE) of 0.42 s and 0.36 s for STm and DT, respectively. On the other hand, EEG-based assessment found a significant effect of condition in TLI, suggesting an increased trend in mental workload during the DT condition, with greater values in DT compared to STc for all the outcomes. The planned inclusion of additional participants is expected to help draw more robust conclusions regarding cortical dynamics in PwMS during DT performance.