Kahlus-NeuroVisual Roadmap#

Need Statement#

A way is needed to structure and quantify brief visual-perceptual episodes in people with prior seizure history without claiming diagnosis.

This branch is future work. It must not support self-triggering, unsupervised photic testing, diagnosis, or treatment.

Safe Initial Scope#

  • retrospective public EEG data review

  • event-schema design for visual-perceptual episode annotation

  • non-diagnostic signal quality and episode-structure metrics

  • clinician/researcher-supervised workflow assumptions for any future real protocol

Dataset Candidates To Investigate#

  • CHB-MIT

  • TUH EEG / TUSZ

  • Siena EEG

  • MOABB

Baselines#

  • seizure/event labels as metadata only when dataset terms allow

  • persistence and spectral baselines

  • ridge/TCN/Transformer EEG baselines

  • identity and leakage probes

Evaluation Criteria#

  • held-out subject/site split where supported

  • no photic stimulation instructions

  • no diagnosis/treatment wording

  • finite metrics

  • explicit quality flags

  • evidence gate blocks epilepsy diagnosis claims

Buildable In 8 Weeks#

  • dataset/license review table

  • safe episode schema draft

  • synthetic annotation fixture

  • no hardware, no app, no triggering task

Long-Term#

  • supervised research annotation tool

  • linkage to v1/v2 representations after strict evidence review

  • prospective protocol only after appropriate review and oversight

Citations To Verify#

Reference

Why It Matters

citation_status

CHB-MIT

public seizure EEG candidate

needs_verification

TUH/TUSZ

public clinical EEG candidate

needs_verification

Siena EEG

seizure EEG candidate

needs_verification

MOABB

non-clinical EEG benchmark tooling

needs_verification