RSWAi_eeg_analysis

You can keep other project helpers (e.g., pre_post_val.m, group_sleep_var.m, task_para4plot.m, groups_indx.m, Fisher_transf.m, donutchart.m, daviolinplot.m) under /helpers as well.

---

Description

This repository provides MATLAB scripts for preprocessing and analyzing high-density EEG sleep recordings from the EGI system.
The pipeline converts raw .mff data into FieldTrip structures, applies artifact cleaning (including ICA), and extracts REM-only segments for advanced analysis.
It further computes power spectral measures, quantifies REM sleep without atonia (RSWAi), and performs group-level statistics and visualizations.
The scripts were used in the study: Beyond RBD: Covert REM sleep abnormalities in Parkinson’s disease (Lanir-Azaria et al., 2023).

Setup

1. Clone/download this repository.
2. Add FieldTrip to your MATLAB path and run ft_defaults.
3. Copy your template files into /templates or update paths in scripts (all paths are written with <YOUR_PATH> placeholders—set them once at the top of each script as instructed in comments).
4. Keep raw data and subject lists under /data (not committed).

---

Input data

• Raw EEG: EGI high-density .mff per subject.
• Sleep staging: CSV/Excel with epoch codes (30-s epochs). Project uses the stage codes:
  • 200 = Wake, 100 = REM, -100 = N1, -200 = N2, -300 = N3.
• Subject list: a MATLAB file (e.g., subjects_files.mat) with fields: mff_file, scoring_csv, subject, group.

---

End-to-end workflow

1) Convert + initial preprocessing (all channels)

Use preper_sleep_data4.m (called by your batch script) to:

• Load .mff, apply band-pass 0.5–45 Hz, remove line noise (50 Hz & harmonics via DFT or notch),
• Resample to 250 Hz,
• Segment into 30-s epochs,
• Align/sanitize staging vector length, and
• Save epoched FieldTrip data (data, epoched_scoring).

Note on line noise: The scripts support both a narrow notch and FieldTrip’s DFT filter (cfg.dftfilter='yes'; cfg.dftfreq=[50 100 150];). Choose one consistently.

2) REM-only artifact cleaning per subject

Run prep_REM_sleep_stages.m to:

• Select REM epochs only.
• Create eyes and brain datasets; eyes electrodes are set to NaN in brain data to preserve indexing.
• Mark bad channels per subject (edit the bad_eyes_chn / bad_brain_chn lists).
• Interactive inspection with ft_databrowser and trial/channel rejection with ft_rejectvisual.
• ICA (ft_componentanalysis, method='runica' by default). Use plot_top_comp.m for component maps; manually list components to remove.
• Reconstruct clean data with ft_rejectcomponent.
• Interpolate bad/missing channels with Fix_bad_Chann (uses ft_channelrepair) and Fix_Miss_Chann.
• Save <subject>_clean_REM.mat and a keep_track struct documenting: removed trials, removed channels, ICA components, etc.

3) Aggregate REM power per group

Use sleep_group_cal.m (REM-only) to:

• Load cleaned subject files per group folder,
• Interpolate/fix channels again for safety,
• Re-reference (avg or as configured), reorder labels if needed,
• Compute power with sleep_frequency_bands.m (bands: All, δ, θ, α, σ, β, γ), and
• Collect into freq_control and freq_PD (per-band cell arrays of FieldTrip freq structures).

4) Group statistics & figures

Run REM_GroupSpectralStats_Main.m to:

• Produce multi-band topographies per group: topo_SpectralBands_3groups_multiFig.m (Control, PD−RSWA, PD+RSWA).
• Compute cluster-based T-maps between groups: Tmaps_3groups_multiFig.m (uses montecarlo_statistics3).
• Plot ROI violin plots and run pairwise tests: daviolinplot_3groups.m / daviolinplot_3groups_singleROI.m.
• Run per-band omnibus ANOVA: bands_ANOVA_3group.m or full permutation ANOVA ANOVA_3group_freq_new.m.

Statistical note. Pairwise tests/visualizations are treated as post-hoc. Use them after an omnibus ANOVA shows a significant group effect, or interpret with appropriate multiplicity control (e.g., BH-FDR with FDR_corr.m).

5) GML task × sleep figures

• GML_Sleep_Correlation_Plots.m – scatter + regression (by subgroup) for any sleep metric vs GML improvement.
• plot_sleep2task_diff_size.m – same but X = REM% (REM/TST×100), with point size by RSWAi via RSWAi_2scattSize.m.
• daviolinplot_RSWAi.m – RSWAi distributions (phasic/tonic/any) for Control vs PD.
• sleep_donut.m – group donut plots of sleep stage composition.

---

Configuration & parameters

• Filtering: 0.5–45 Hz band-pass. Add DFT notch at 50/100/150 Hz as needed.
• Epoching: 30-s (aligned to staging). Ensure scoring length matches data length; script pads/trims accordingly.
• Resampling: to 250 Hz if original is 1000 Hz.
• Stage codes: 200=Wake, 100=REM, -100=N1, -200=N2, -300=N3.
• Frequency bands: [0.5–40/45], δ [0.5–4], θ [4–8], α [8–12], σ [12–15], β [15–25], γ [25–45] (configurable).
• Statistics: BH-FDR via FDR_corr.m; permutation ANOVA via ANOVA_3group_freq_new.m.

---

Manual steps & documentation

• Bad channels: set bad_eyes_chn, bad_brain_chn per subject before REM cleaning.
• Visual rejection: use ft_rejectvisual in prep_REM_sleep_stages.m (keeps NaNs to preserve indexing where necessary).
• ICA components: decide visually using plot_top_comp.m; list components to remove in the script. ECG-like components can also be saved separately.
• Provenance: the keep_track struct records removed channels, trials, and components for reporting.

---

Troubleshooting

• Line noise persists: prefer FieldTrip’s dftfilter with dftfreq=[50 100 150]. Avoid stacking notch + DFT simultaneously.
• Label mismatches (E257/VREF): use Fix_bad_Chann/Fix_Miss_Chann and provided templates to normalize labels.
• ICA fails or components look odd: check referencing and that only brain channels (not eye channels) drive the ICA input.
• Missing functions: some plotting/stat helpers are third-party (daviolinplot, donutchart). Add them to the MATLAB path.

---

How to cite

Lanir-Azaria S, Nir Y, Tauman R, Zitser J, Giladi N. (2023). Beyond RBD: Covert REM sleep abnormalities in Parkinson’s disease. npj Parkinson’s Disease.

---

Contact

Questions or issues? Please open an issue or contact Saar Lanir-Azaria.

saarlan530@gmail.com