The unmet need that justifies opening a skull
You reach for an implant only when everything gentler has failed. In major depression, roughly a third of patients remain ill after adequate trials of several drugs and psychotherapy — the state clinicians call treatment-resistant depression. A similar tail of severe, disabling obsessive-compulsive disorder resists every front-line therapy. For that tail, deep brain stimulation is offered not as an enhancement but as a therapy of last resort, and OCD DBS already carries a humanitarian-device authorisation. This is the population psychiatric BCI serves: not the average patient, but the one for whom nothing else worked.
The first generation, and its honest setbacks
Open-label pilot studies of DBS for depression were genuinely exciting. Stimulating the subcallosal cingulate (Area 25) — a hub over-active in depression — or the ventral capsule / ventral striatum produced striking recoveries in some treatment-resistant patients. It looked like psychiatry had found its pacemaker.
Then the blinded, randomised pivotal trials fell short. The two large industry-sponsored studies — one targeting the subcallosal cingulate, one targeting the ventral capsule/ventral striatum — did not beat sham on their primary endpoints and were stopped early for futility. The lesson was not that DBS does nothing; it was that a fixed, open-loop dose delivered to an anatomically-fixed target, evaluated against the huge expectation effects of brain surgery, is not reliably distinguishable from placebo. That single sentence motivates almost everything in this track.
Why reading a mood is uniquely hard
Motor BCI has an enormous advantage psychiatry lacks: a fast, objective, second-by-second ground truth — the cursor either hit the target or it did not. Mood has none of this. A neural signature of mood must be learned against labels that are sparse (a few self-reports a day), subjective, slow (mood evolves over hours to days), and non-stationary (today's biomarker may drift by next month). On top of that, the read/write asymmetry bites hard: even if you could read the state, the stimulation you write back changes it, so cause and effect are tangled inside the loop.
The whole shift in one line. Open-loop DBS delivers a constant dose u₀. Closed-loop DBS extracts a biomarker bₜ from the neural signal yₜ via a decoder f_θ, then a policy π sets the stimulation uₜ from that state. Everything downstream is engineering this map honestly.
The shape of the frontier
The frontier therefore has two coupled halves, and the remaining guides follow them: where to intervene (networks, not just nuclei), and when and how much to intervene (adaptive, biomarker-driven stimulation). Wrapping both is a move from population-average protocols to per-patient personalisation — because in a disease with no shared read-out, the biomarker and the target may both be individual. That is the honest promise of closed-loop DBS for depression: not magic, but a way to make the intervention answerable to the brain it is treating.