Code to accompany the paper High frequency oscillations measured with optically pumped magnetometers from the human retina.
The study records the flash-evoked electroretinogram (ERG) with optically pumped magnetometers (OPMs) and electrodes simultaneously. Fourteen OPMs were mounted in front of the eye, alongside two ERG electrodes (above and below the eye). The analyses compare the low-frequency response (a-/b-wave) and the oscillatory potentials (75–170 Hz) between the two modalities.
Data are not included in this repository.
- Python ≥ 3.10:
pip install -r requirements.txt - MATLAB (mutual-information analysis only), with
mutualinformationx.mfrom M. X. Cohen, Analyzing Neural Time Series Data (MIT Press, 2014) on the path.
Set OPM_ERG_DATA_DIR to a folder with one sub-folder per participant:
<data dir>/
PO13/
PO13_raw.fif OPM recording (trigger channel Input-1)
PO13000.0100.flt(.hdr) EEG: 4 interleaved float32 channels (E01-E03, analog trigger)
PO10/ PO10-PO12: EEG stored as one file per channel
PO10.C01_0000_CH368.$$$ ...converted once with convert_split_eeg.py
Participants PO10–PO19 were recorded. PO12 is excluded from group analyses because the OPMs were not initialised properly in that session.
Run the Python scripts from python/. Results and figures go to ./results
(change with --out); derivative data files go to the participant folders.
| Step | Script | Output |
|---|---|---|
| 1 | convert_split_eeg.py |
EEG .fif for PO10–PO12 |
| 2 | preprocess.py --band 110 145 |
evoked files (0.5–45 Hz, 75–145 Hz, 110–145 Hz, 130–145 Hz envelope) and 1 Hz high-pass epochs |
| 3 | compute_itc.py |
Morlet ITC and power, 110–170 Hz (20 log-spaced frequencies, n_cycles = f/10) |
| 3b | compute_itc.py --fmax 145 --n-freqs 10 --divisor 9 --tag 9cycles |
ITC 110–145 Hz, used by fig_itc_grand_average.py |
| 3c | compute_itc.py --fmax 300 --n-freqs 35 --divisor 15 --tag 15cycles6 |
ITC 110–300 Hz (35 log-spaced frequencies, n_cycles = f/15, 7.3–20 cycles); Figure 5b with fig_itc_grand_average.py --fmax 300 --tag 15cycles6 |
| 4 | itc_stats.py |
Group statistics on the ITC (Figure 5) |
| 5 | export_mat.py |
.mat files for MATLAB |
| 6 | matlab/mi_window.m, plot_mi_sensitivity_map.m |
MI between each sensor and the ERG, -10 to 120 ms (Figure 6) |
| 7 | matlab/mi_itc_over_time.m |
MI of ITC time courses over time and frequency (Figure 7) |
| – | matlab/mi_over_time.m |
MI of the evoked envelope in a sliding window (supplementary) |
Figures: fig_evoked_low.py (Figure 2), fig_evoked_high.py (Figure 3B),
fig_best_opm_individual.py (Figure 4B), fig_itc_grand_average.py,
fig_amplitude_spectrum.py (Figure 8). Supplementary comparisons:
snr_compare.py, itpc_compare.py.
Both systems sample at 1 kHz and receive the same flash trigger. Data are
epoched from −200 to 300 ms on each system's own trigger and merged. The first
10 flashes of each run are discarded, then epochs rejected by visual inspection
(rejected_epochs.py). Filters are applied to the continuous data before epoching.
For every channel and frequency, a within-participant Mann–Whitney U compares each time point with the pre-stimulus period; the U values are signed by whether the ITC lies above the baseline median. Across participants, the signed values at each time point (−50 to 100 ms) are compared with all baseline values (−50 to 0 ms). p-values are FDR-corrected over time.
Histogram-based MI between each sensor and the under-eye ERG, on the 130–145 Hz
envelope of the evoked response, −10 to 120 ms (131 samples). The bin count is the
rounded mean of Freedman–Diaconis estimates. Significance is a z-score against
surrogates in which the ERG bin sequence is split at a random point and its
segments swapped/reversed (2000 permutations per participant, 5000 for the group).
The group result is the MI of the grand-average envelopes. The permutations are
seeded (mi_window(signal, seed), default seed 1), so Figure 6 is reproducible.
Figure 7 uses the same estimator on ITC magnitude time courses (4 bins, 27-sample
sliding window, 1 ms steps): MI in bits per participant, averaged over the nine
participants and plotted from −10 to 120 ms. mi_itc_over_time(nPermutations) with
nPermutations > 0 adds permutation statistics (rank-sum across participants against
the −50–0 ms baseline, and the MI of the grand-average ITC); these are not in the paper.
Every evoked/epochs file has 16 channels: 14 OPMs (index 0–13), ERG above the eye
(14) and ERG below the eye (15). "OPM 13" (index 12) is the sensor with the
clearest response. Layout settings are in python/config.py and matlab/study_config.m.