Auto-Adaptive Model for Longitudinal Motor Imagery Decoding in Amyotrophic Lateral Sclerosis

Amyotrophic Lateral Sclerosis (ALS) has been a grossly misrepresented end user group when developing coadaptive algorithms for Brain Computer Interfaces (BCI). Researchers have credited this issue to the difficulty of progressing disease in patients with ALS. This non-stationarity reduces accuracy over time. This paper introduces an online model, usable for a BCI using ALS patients data. The automatic coadaptive model effectively decodes 3 class motor imagery (MI) of the left, right hand and rest while adapting to address non-stationarities of Electroencephalography (EEG) over time caused by various factors over the study duration. Adapting Filter bank Common Spatial Pattern (FBCSP) algorithm, where we show it could enable above 70% detection of hand MI in ALS end users longitudinally, previously lacking evidence. The evaluation results demonstrate that the model achieves average accuracies of 72.6% over a 1–2 month period of usage involving 8 ALS patients. This work shows the first auto-adaptive model with ALS patient EEG data providing a stronger incentive for further investigation by setting benchmark models on longitudinal datasets contributing to the solution of multiple challenges in this field.