Multimodal data for Alzheimer's model

# Data `oasis_longitudinal.csv` here is a **synthetic** sample (40 fake subjects, 82 visits) that mimics the schema of the OASIS-2 longitudinal dataset, so the demo runs out of the box without any data agreement. To use real data, download the OASIS-2 longitudinal CSV from [oasis-brains.org](https://www.oasis-brains.org/) and replace this file — same column names, no code changes needed. ## Phase-2 imaging data `download_oasis.py` in this directory fetches the OASIS **imaging** dataset from Kaggle (`ninadaithal/imagesoasis`) for the phase-2 MRI severity classifier ([src/alz/imaging.py](../src/alz/imaging.py)). It lands at `data/imagesoasis/versions/1/Data/<class folders>` (kagglehub's cache layout) and is unrelated to the tabular workflow above -- running it does not affect `oasis_longitudinal.csv` or `train.py`. Train with `python train_mri.py` (see [README.md](../README.md)). `imagesoasis/versions/1/Data/` holds one folder per class, each full of `.jpg` MRI slices: `Mild Dementia`, `Moderate Dementia`, `Non Demented` (a `Very mild Dementia` class also exists upstream on Kaggle if you re-download). ## Phase-3 EEG data `ds004504/` is the OpenNeuro BIDS dataset "A dataset of EEG recordings from Alzheimer's disease, Frontotemporal Dementia and Healthy subjects" (Miltiadous et al.; originally `github.com/OpenNeuroDatasets/ds004504`). It feeds the phase-3 EEG confirmation classifier ([src/alz/eeg.py](../src/alz/eeg.py)), which extracts relative band-power features (delta/theta/alpha/beta/gamma) and fits a logistic regression. 88 subjects: 36 Alzheimer's (AD), 23 Frontotemporal Dementia (FTD), 29 healthy controls (CN). Each `sub-0XX/eeg/` holds the raw resting-state eyes-closed recording (`.set`, Nihon Kohden 19-channel montage, 500 Hz); `derivatives/sub-0XX/eeg/` holds the same recording after band-pass filtering, ASR artifact correction, and ICA cleaning. `participants.tsv` columns: `participant_id, Gender, Age, Group, MMSE`, where `Group` is `A` (AD), `F` (FTD), or `C` (healthy control). ## Raw OASIS-1 MRI (supplementary) `OASIS1_raw/` is the original OASIS-1 cross-sectional MRI release from [oasis-brains.org](https://www.oasis-brains.org/), split across two discs: `oasis_cross-sectional_disc1/disc1/` (39 subjects) and `oasis_cross-sectional_disc2/disc2/` (38 subjects), plus a demographics spreadsheet (`oasis_cross-sectional-*.xlsx`). Each `OAS1_XXXX_MR1/` subject folder has `RAW/` (unprocessed scans), `PROCESSED/MPRAGE/` (registered/atlas-aligned volumes), and `FSL_SEG/` (tissue segmentation), all in Analyze format (`.hdr`/`.img`). This is raw source data -- no script in this repo currently reads it; it's kept for reference and future MRI work beyond the Kaggle `imagesoasis` slices used today. ## Expected columns `Subject ID, MRI ID, Group, Visit, MR Delay, M/F, Hand, Age, EDUC, SES, MMSE, CDR, eTIV, nWBV, ASF` - `Group`: `Nondemented` / `Demented` / `Converted` — the prediction target. - `SES`, `MMSE` commonly have missing values; `Subject ID`, `MRI ID`, `Hand`, `MR Delay`, `eTIV`, `CDR` are dropped or unused as predictors (see [src/alz/data.py](../src/alz/data.py)), following the preprocessing in [vendor/alzheimers-disease-prediction/dimentia_pred.ipynb](../vendor/alzheimers-disease-prediction/dimentia_pred.ipynb).