Depth-of-Anesthesia Estimation from EEG — Project Root

A two-phase research project on subject-independent Depth-of-Anesthesia (DoA) estimation from single-channel EEG on the VitalDB cohort.

Target venue: IEEE Transactions (TBME / TNSRE / JBHI) Author: Fahad Ahmed Current phase: Transitioning from v2 (mini-project, done) to v3 (journal paper, 12 weeks)

---

Folder map

Folder	Purpose
01_docs/	Planning documents. Current roadmap lives here. Older plans are in archive/.
02_literature/	Reference papers, organised by topic (baseline, deep_learning, nirs, reference).
03_data/vitaldb_raw/	Raw EEG + BIS recordings. 24 cases at the moment; v3 Week 1 expands this to ≥100.
04_pipeline_v2/	The mini-project codebase. Frozen. Git repository intact. Reproduces the v2 results: baseline RMSE 10.65, LOPO RMSE 11.53 ± 2.13.
05_pipeline_v3/	The new pipeline scaffold. Empty skeleton matching the roadmap's module layout. Week 1 of the v3 plan starts here.

Where to start

1. Read 01_docs/DoA_Research_Roadmap_v3.docx end to end — it is the single source of truth for the plan, the architecture, and the 12-week schedule.
2. Section 13 of the roadmap is a 48-hour unblock checklist for Week 1.
3. The v2 results that v3 must beat are summarised in Section 2 of the roadmap and reproducible from 04_pipeline_v2/.

Reproducing v2 (sanity check)

cd 04_pipeline_v2
pip install -r requirements.txt       # if not yet installed
python scripts/main_pipeline.py       # runs inspect → preprocess → features → baseline → LOPO → results

Configuration automatically resolves 03_data/vitaldb_raw/ after the v3 reorg, with the old Dataset/ path kept as a fallback.

Headline v2 results (from 04_pipeline_v2/outputs/results/)

Metric	Value	Comparison
Best baseline RMSE (80/20)	10.65	Beats Rani 2020 target of 11.73
Best baseline MAE	6.97	—
Best baseline Pearson r	0.777	—
Best LOPO RMSE	11.53 ± 2.13	Cross-patient validated
Best LOPO Pearson r	0.700	—