Longitudinal study of gesture decoding in a clinical trial participant with ALS.

Brain-computer interfaces (BCIs) have the potential to preserve or restore communication and device control in people with paralysis from a variety of causes. For people living with amyotrophic lateral sclerosis (ALS), however, the progressive loss of cortical motor neurons could theoretically pose a challenge to the stability of BCI performance. Here we tested the stability of gesture decoding with a chronic electrocorticographic (ECoG) BCI in a man living with ALS and participating in a clinical trial (ClinicalTrials.gov, NCT03567213). We evaluated offline decoding performance of attempted gestures over two periods: a 5-week period beginning roughly 2 years post-implant and a 6-week period ending roughly 5 months later. Decoder sensitivity was high in both periods (90 - 98%), while classification accuracy was 37 - 68% in the first period and worsened to 23 - 39% in the second. We investigated multiple frequency bands that were used as model features in both periods, and we observed reductions in high gamma band power (70 - 110 Hz) and between-class separation during the second period compared to the first. Over the 5-month period motor function did not appreciably decline. These results, albeit preliminary, suggest that declines in the neural population responses that drive ECoG BCI performance can occur without overt signs of disease progression in people living with ALS, and could serve as a biomarker for disease progression in the future.