Low-cost cognitive screening: consumer EEG, a marker-annotated film and age norms
Six directions around cognitive-aging screening on consumer EEG: in-browser event marking, ERP and inter-subject correlation against age norms, remote hyperscanning of dyads, objective pharmaco-EEG endpoints, the evidential value of a record against the right to erasure, and comparability of data across device updates — with the status of each direction stated openly.
Koch Laboratory — low-cost cognitive screening: consumer EEG, a marker-annotated film, age norms
This direction asks how much clinically useful information about cognitive aging can be obtained from four channels of consumer EEG (dry electrodes, Bluetooth transmission) — when the stimulus is not a tedious test but a purpose-built film with time markers for events, and the result is referenced to an age norm and to the person’s own history. The foundation joins two established paradigms: event-related potentials (ERP — the latency of the P300 component lengthens with age by ~1–2 ms/year, and more markedly in cognitive impairment) and inter-subject correlation (ISC — the synchronization of EEG responses across viewers of the same film measures attention and declines with age). The market reference point is an FDA-cleared ERP system using audio stimuli and seven electrodes; this program’s niche is screening an order of magnitude cheaper, short and engaging (a film instead of an oddball), scalable through the browser.
Method. As in the rest of the lab: problem → hypothesis → falsifiable criterion → method → result with boundary conditions. The boundary of responsibility is part of the design, not decoration: the tool provides information supporting a physician’s decision and never adjudicates; the medical-device path (MDR) is deliberately a separate, later stage.
Publication note (IP). For directions with filing potential we publish the problem, the state of the art and the success criterion — not the solution’s construction. Where “Construction: withheld” appears, the technical detail remains unpublished.
1. Direction 1 — marking stimulus events against EEG in the browser
Research question. Can a platform running entirely in the browser (Web Bluetooth, no installation) mark film events precisely enough for ERP averaging, given that Bluetooth transmission introduces a variable lag of tens of milliseconds while the requirement is ~10 ms? Why it is hard. ERP emerges from noise only by averaging over precisely marked repetitions — variable lag smears the components. Proven laboratory solutions (a shared clock and timestamp correction on the desktop side) do not transfer directly to the browser, which adds its own buffering layers. State of the art (published). Validated desktop pipelines with timestamp correction (the Krigolson group); libraries for streaming consumer EEG in the browser; hardware synchronization checks (photodiode, audio marker) as the reference method. Success criterion. A visible averaged event-related potential after ≥30 repetitions of an event in a browser recording; timing agreement with a parallel laboratory pipeline. Construction (time-correction mechanism, handling of hardware and firmware versions): withheld. Status: prototype in progress — the in-browser headband connection, live view of the four channels, electrode-quality indicators, and recording with markers and firmware version all work; the ERP criterion is not met.
2. Direction 2 — aging screening: ERP + ISC against norms and one’s own history
Research question. Do ERP component latencies and ISC strength, measured on four channels during a marker-annotated film, position a person against an age-matched normative group (percentile) and detect change against their own baseline? Why it is hard. Four dry electrodes are a fraction of a clinical montage; a film is a richer but less controlled stimulus than the classical oddball; and longitudinal analysis is worthless without strict stimulus versioning and complete session metadata — responses to two different film versions are not directly comparable. The program’s largest investment is not code but data: nobody holds a normative cohort for older age groups on this hardware. State of the art (published). ERP assessment systems with audio stimuli (FDA clearance); validation of the consumer headband for ERP research and mass measurement of N200/P300 against age; age-dependent variability of responses to naturalistic video; ISC as a measure of attention. Success criterion (two-part). (a) Replication of known age effects (latency lengthening, ISC decline) on the platform itself; (b) longitudinal sensitivity: detecting change against one’s own baseline with a fixed stimulus version. Both open. Status: design work; the lab’s own pilot recordings exist but without event markers — openly stated as the main gap (without labels there is no ERP analysis and no supervised learning); the baseline data-collection protocol is defined, and the data schema is designed from the first record for multi-year comparability.
3. Direction 3 — remote hyperscanning: inter-brain synchrony in dyads
Research question. Does the emotional closeness of a dyad raise inter-brain synchrony (ISC) while both members simultaneously watch the same marker-annotated film in two locations? Why it is hard. Remote hyperscanning multiplies artifact sources (two devices, two links, two clocks), and small effects in two-person data are easy to “manufacture” through analysis — without hard controls the result is unpublishable. State of the art (published). Hyperscanning and interpersonal neural synchronization (INS) are established currents of social neuroscience; the remote variant went mainstream after the pandemic; ISC with naturalistic stimuli is well described. Success criterion. Preregistration of hypotheses and analyses; an effect present in real dyads and vanishing in sham pairs (combinations of people who were never connected) and in surrogate data (circular shifts); permutation tests with correction for multiple comparisons. Status: design work — the shared-session architecture is designed, implementation not started.
4. Direction 4 — objective pharmaco-EEG endpoints
Research question. Is longitudinal measurement against one’s own baseline (a session before administration → after → follow-up) on a cheap, scalable platform suitable as an objective endpoint in studies of centrally acting substances — where today measurement relies mostly on questionnaires? Why it is hard. Four channels will detect strong signatures (sedation) easily; subtle effects demand group statistics over many participants — which is exactly the cheap scaling that is the platform’s hypothesis, not its proven property. State of the art (published). Pharmaco-EEG as an established discipline with known EEG signatures of drug classes; a growing research line on prosocial substances whose effects have so far been measured subjectively. Success criterion. First, reproduction of known strong signatures on the platform itself; only then subtle endpoints. Boundary of responsibility (published as a principle). The platform is a measuring instrument: it does not dose, recommend or recruit; substance studies run under the ethical oversight and responsibility of the sponsor. Status: future direction, design work.
5. Direction 5 — evidential value of the record against the right to erasure
Research question. Can a session record be immutable and time-anchored enough to serve as evidentiary material for a study, and still be subject to the participant’s effective right to have their health data erased? Why it is hard. The two requirements contradict each other directly. An audit log carries evidential weight only as long as nothing disappears from it; data protection law grants the right to withdraw consent and demand erasure of special-category data — and an EEG recording is one. Partial answers such as pseudonymisation or anchoring of checksums serve one side and weaken the other. State of the art (published). RFC 3161 time stamps, append-only logs, pseudonymisation and the requirements on the reliability of clinical study data (ALCOA+) are each established on their own; for their combination in a system that must simultaneously honour an erasure request and remain verifiable to an outside auditor, no established pattern exists. Success criterion. A procedure in which withdrawal permanently removes access to the measurement data while an independent auditor can still show that the record existed at a given time and was not altered — together with an explicit statement of what can no longer be shown once erasure has taken place. Construction (how the contradiction is resolved): withheld. Status: design work; the data model is implemented — consents kept as an event table with the version of the text and the moment of withdrawal, an append-only operations log, exclusion of a session from the norm as a reasoned flag rather than a deletion. Anchoring and the resolution of the conflict with the right to erasure remain open.
6. Direction 6 — comparability of data across device updates
Research question. Can the comparability of a normative cohort collected over years be preserved when access to the raw signal rests on a reverse-engineered protocol and the manufacturer changes device firmware without publishing a list of changes? Why it is hard. The criterion of correctness is not “the connection works” but “the data remain comparable with a norm collected on a different version”. The update is one-way and runs through the manufacturer’s own application, so a version cannot be rolled back; and a change in the device’s internal pre-processing stays invisible from the outside — the signal continues to look plausible. State of the art (published). Communities around consumer EEG have maintained the reverse-engineered protocol for years, and supporting several device generations is ordinary engineering practice — in both cases, however, the measure of success is that a connection is established, not that measurements remain comparable across versions. Success criterion. A reference test that decides, for any two versions, whether recordings may be treated as interchangeable within the same normative cohort; together with a version status visible before a session starts — tested, experimental or blocked. Construction (compatibility matrix and version qualification procedure): withheld. Status: design work; the zero point of the matrix is in hand (a device not updated for over a year), the protocol client reads model and firmware version from the device, the reference test has not yet been devised.
Program status: active; the browser acquisition prototype works, all measurement criteria remain open; the next stage is meeting the ERP criterion of Direction 1 and the first recordings with event markers.