Stable speech BCI performance during slow progression of ALS: A longitudinal ECoG study

Area of Science:

• Neuroscience
• Biomedical Engineering
• Neurology

Background:

• Electrocorticographic (ECoG) speech brain-computer interfaces (BCIs) offer potential communication restoration for amyotrophic lateral sclerosis (ALS).
• Long-term stability of neural signals and decoding performance during ALS progression is not well understood.
• Investigating high-gamma (HG) activity changes over time and their impact on speech decoding is crucial.

Purpose of the Study:

• To track ECoG speech signal characteristics and decoding performance over 25 months in an ALS participant.
• To determine how high-gamma (HG) activity changes over time during disease progression.
• To assess the impact of these neural changes on offline speech decoding accuracy and durability.

Main Methods:

• Implanted ECoG grids over left sensorimotor cortex (SMC) in an ALS participant.
• Recorded ECoG and audio during an overt syllable-repetition task over 25 months.
• Quantified HG activation ratio (ActR), signal-to-noise ratio (SNR), and acoustic features; assessed offline EEGNet decoders.

Main Results:

• Speech acoustics showed reduced vowel space area (tVSA) over time, indicating mild intelligibility decline.
• Neural metrics (ActR, SNR) exhibited a biphasic trajectory: initial increase followed by SNR decline.
• Decoders trained on stabilized data (months 7-11) showed robust performance on later data, with no significant temporal decline.

Conclusions:

• Speech-related HG features show initial instability followed by gradual SNR reduction, potentially reflecting ALS progression.
• Models trained after signal stabilization generalize robustly, supporting durable ECoG speech BCIs without frequent recalibration.
• Findings support adaptive calibration algorithms leveraging stable spatial representations in ventral SMC for future BCIs.