Intracranial mood decoding

Using machine-learning models to predict a person's self-reported mood or symptom severity from intracranial EEG recorded chronically or during inpatient monitoring. Because mood is sampled sparsely — a handful of self-reports per day — decoders must learn from limited, noisy labels and cope with drift over days; approaches range from regularized linear models on band-power features to latent-state dynamical models.

Decoding is the read side of a closed loop: a validated decoder both identifies which biomarker to target and supplies the trigger signal that gates stimulation. Reported accuracies are encouraging within a subject but do not yet transfer across people, and even the best decoder is bounded by the reliability of the self-reports it is trained against.