Package 'mrpheus'

Description

Designs a 4th-order Butterworth bandpass filter and applies it zero-phase using gsignal::filtfilt(). Handles edge cases gracefully:

Usage

bandpass_filter(signal, sr, low_hz, high_hz)

Arguments

Details

• low_hz <= 0 → low-pass at high_hz

• high_hz >= Nyquist → high-pass at low_hz

• Otherwise → bandpass between low_hz and high_hz

The high_hz cutoff is clamped to 99 % of Nyquist to avoid instability. Because this uses a short IIR filter, it is safe on long recordings.

Value

Filtered numeric vector, same length as signal.

Examples

## Not run: 
# Standard EEG bandpass
filtered <- bandpass_filter(eeg_signal, sr = 256, low_hz = 0.3, high_hz = 35)

# High-pass only (baseline drift removal)
hp <- bandpass_filter(eeg_signal, sr = 256, low_hz = 0.1, high_hz = Inf)

## End(Not run)

Description

Calculates the AHI from respiratory event data and total sleep time.

Usage

compute_ahi(events, tst_hours)

Arguments

Value

Numeric scalar. AHI (events per hour).

Description

Estimates power spectral density (PSD) using Welch's method and integrates power within standard EEG frequency bands for each epoch and channel.

Usage

compute_band_power(
  psg,
  channels = NULL,
  bands = list(delta = c(0.5, 4), theta = c(4, 8), alpha = c(8, 12), sigma = c(12, 16),
    beta = c(16, 30), gamma = c(30, 45)),
  relative = FALSE,
  win_sec = 4,
  noverlap = 200L,
  nfft = 1024L
)

Arguments

list(delta = c(0.5, 4), theta = c(4, 8), alpha = c(8, 12),
     sigma = c(12, 16), beta = c(16, 30), gamma = c(30, 45))

Details

The Welch implementation matches scipy.signal.welch with YASA's default parameters (win_sec = 4, noverlap = 200, nfft = 1024, Hann window, constant detrend, mean averaging). Relative power matches YASA's relative = True behaviour: each band is divided by the sum of all band powers (not the total PSD integral).

Value

A tibble with columns epoch, channel, one column per named band, and total_power (sum of all band powers, in V^2/Hz units before any relative scaling).

Examples

## Not run: 
bp <- compute_band_power(psg)
bp <- compute_band_power(psg, channels = "EEG Fpz-Cz", relative = TRUE)

# Custom bands
bp <- compute_band_power(
  psg,
  bands = list(slow = c(0.5, 1), delta = c(1, 4), theta = c(4, 8))
)

## End(Not run)

Description

Converts R-peak sample indices (from detect_qrs()) into a sample-by-sample instantaneous heart rate trace by filling each inter-beat interval with the constant HR derived from that interval's length.

Usage

compute_hr_signal(qrs)

Arguments

Value

Numeric vector of length max(qrs$qrs_i). Each sample contains the instantaneous heart rate in beats per minute for the inter-beat interval it falls within. Samples before the first detected R-peak are 0.

See Also

detect_qrs(), compute_hrv_sleep()

Examples

## Not run: 
rec <- read_philips_physlog("sub-01_physlog.log")
qrs <- detect_qrs(rec$C[, "v1raw"], fs = rec$HDR$sfreq)
hr  <- compute_hr_signal(qrs)

plot(seq_along(hr) / rec$HDR$sfreq, hr, type = "l",
     xlab = "Time (s)", ylab = "HR (bpm)")

## End(Not run)

Description

Extracts R-peak positions from the ECG channel, computes standard time- and frequency-domain heart rate variability (HRV) metrics, and stratifies results by sleep stage.

Usage

compute_hrv_sleep(
  psg,
  staging = NULL,
  ecg_channel = NULL,
  min_rr_ms = 300,
  max_rr_ms = 2000
)

Arguments

Value

A tibble with one row per sleep stage (or one row if staging is NULL):

stage

Character. AASM stage or "ALL".

n_epochs

Integer. Number of epochs in this stage.

mean_rr_ms

Numeric. Mean RR interval (ms).

sdnn_ms

Numeric. SDNN — SD of all NN intervals.

rmssd_ms

Numeric. RMSSD — root mean square of successive differences.

lf_power

Numeric. LF band power (0.04–0.15 Hz).

hf_power

Numeric. HF band power (0.15–0.4 Hz).

lf_hf_ratio

Numeric. LF/HF ratio.

Description

Calculates the ODI (number of >= 3% SpO2 desaturations per hour of sleep) from the SpO2 channel.

Usage

compute_odi(psg, spo2_channel, tst_hours, threshold = 3)

Arguments

Value

Numeric scalar. ODI (desaturations per hour).

Description

Detects slow oscillations (SOs) in EEG channels using a zero-crossing approach in the delta band (0.5-2 Hz), following the algorithm described in Molle et al. (2002) and implemented in YASA (Vallat & Walker, 2021).

Usage

compute_slow_oscillations(
  psg,
  channel = NULL,
  stages = NULL,
  freq_so = c(0.5, 2),
  amp_ptp_threshold_uv = c(75, 500),
  duration_pos_s = c(0.1, 1),
  duration_neg_s = c(0.1, 1.5)
)

Arguments

Value

A tibble with one row per detected slow oscillation:

epoch

Integer.

start_s

Numeric. Onset (s) relative to epoch start.

end_s

Numeric. Offset (s) relative to epoch start.

duration_s

Numeric.

neg_peak_uv

Numeric. Negative peak amplitude (uV).

pos_peak_uv

Numeric. Positive peak amplitude (uV).

ptp_uv

Numeric. Peak-to-peak amplitude (uV).

channel

Character.

References

Molle, M., Marshall, L., Gais, S., & Born, J. (2002). Grouping of spindle activity during slow oscillations in human non-rapid eye movement sleep. Journal of Neuroscience, 22(24), 10941-10947.

Vallat, R., & Walker, M. P. (2021). An open-source, high-performance tool for automated sleep staging. eLife, 10, e70092. doi:10.7554/eLife.70092

Description

Generates a short-time Fourier transform (STFT) based spectrogram for a single PSG channel across all epochs. Returns power (µV²/Hz) as a matrix with time (epochs) on rows and frequency on columns.

Usage

compute_spectrogram(
  psg,
  channel = NULL,
  freq_range = c(0, 40),
  window_s = 2,
  overlap = 0.5,
  db = TRUE
)

Arguments

Value

A list of class mrpheus_spectrogram with:

power

Numeric matrix. Rows = epochs, columns = frequency bins.

freqs

Numeric vector. Frequency bin centres (Hz).

epochs

Integer vector. Epoch indices.

channel

Character. The channel label.

db

Logical. Whether power is in dB.

Description

Detects sleep spindles in EEG channels using a band-pass / RMS envelope approach, closely following the algorithm in Lacourse et al. (2019) and implemented in YASA (Vallat & Walker, 2021). Spindles are identified as transient bursts of 11-16 Hz activity during NREM sleep.

Usage

compute_spindles(
  psg,
  channel = NULL,
  stages = NULL,
  freq_spindle = c(11, 16),
  rms_window_s = 0.3,
  min_duration_s = 0.5,
  max_duration_s = 3,
  threshold_sd = 1.5
)

Arguments

Value

A tibble with one row per detected spindle:

epoch

Integer. Epoch in which the spindle was detected.

start_s

Numeric. Spindle onset relative to epoch start (seconds).

end_s

Numeric. Spindle offset relative to epoch start (seconds).

duration_s

Numeric. Spindle duration in seconds.

peak_freq_hz

Numeric. Peak frequency within the spindle.

rms_uv

Numeric. Mean RMS amplitude within the spindle (uV).

channel

Character. Channel label.

References

Lacourse, K., Yetton, B., Mednick, S., & Bhatt, D. L. (2019). Massive online sleep staging: A polysomnography data repository. Journal of Sleep Research.

Vallat, R., & Walker, M. P. (2021). An open-source, high-performance tool for automated sleep staging. eLife, 10, e70092. doi:10.7554/eLife.70092

Description

Applies a sliding window over the scored epochs, concatenates the signal within each window, estimates the Welch PSD averaged across channels, and integrates power within standard EEG frequency bands. Returns a long-format tibble suitable for plotting temporal power dynamics across the night.

Usage

compute_temporal_bandpower(
  psg,
  hypno,
  channels = NULL,
  window_epochs = 10L,
  step_epochs = NULL,
  bands = list(delta = c(0.5, 4), theta = c(4, 8), alpha = c(8, 12), sigma = c(12, 16),
    beta = c(16, 30), gamma = c(30, 45)),
  win_sec = 4,
  noverlap = NULL,
  nfft = NULL
)

Arguments

list(delta = c(0.5, 4), theta = c(4, 8), alpha = c(8, 12),
     sigma = c(12, 16), beta = c(16, 30), gamma = c(30, 45))

Details

Mirrors calculate_temporal_bandpower() from the companion Python notebook: windows step by 50 % of their length (default), Welch overlap defaults to half the window length in samples (proportional to sample rate), and relative power is the share of each band in the total band-summed power.

Value

A tibble with columns:

time_hours

Numeric. Start time of the window in hours from recording onset.

epoch_start

Integer. First epoch index in the window (1-based).

epoch_end

Integer. Last epoch index in the window (1-based).

dominant_stage

Integer. Most common stage code across the window (artefact epochs excluded). NA if all epochs are artefacts.

band

Character. Band name.

power

Numeric. Mean band power across channels (V^2/Hz).

relative_power

Numeric. Band power as a fraction of total band-summed power at this time point.

Examples

## Not run: 
rec   <- read_edf("SC4001E0-PSG.edf")
psg   <- prepare_psg(rec) |> preprocess_psg()
hypno <- stage_epochs(psg)$stage

tbp <- compute_temporal_bandpower(psg, hypno)

# Plot delta power across the night
library(ggplot2)
tbp |>
  dplyr::filter(band == "delta") |>
  ggplot(aes(time_hours, relative_power)) +
  geom_line() +
  labs(x = "Time (hours)", y = "Relative delta power")

## End(Not run)

Description

Decomposes the EEG into independent components via FastICA, identifies components whose time courses correlate with the EOG channels above threshold, subtracts their contribution from the EEG, and returns a cleaned mrpheus_psg. Mirrors MNE's ICA EOG rejection pipeline with default settings matching the companion Python notebook (n_components = min(6, n_eeg), threshold = 0.35).

Usage

correct_eog_ica(
  psg,
  eog_channels = NULL,
  eeg_channels = NULL,
  n_components = NULL,
  threshold = 0.35,
  fun = "logcosh",
  verbose = TRUE
)

Arguments

Details

Note: FastICA (this implementation) and MNE's default InfoMax ICA find different decompositions; the identified components and final signal will therefore differ from MNE output, but artifact removal performance is comparable.

Value

A new mrpheus_psg with cleaned EEG signals and re-segmented epochs. EOG and other channels are unchanged.

Examples

## Not run: 
rec   <- read_edf("psg.edf")
psg   <- prepare_psg(rec) |> preprocess_psg()
clean <- correct_eog_ica(psg)

# Stricter threshold
clean <- correct_eog_ica(psg, threshold = 0.5)

## End(Not run)

Description

For each EEG channel, fits a multiple linear regression on all EOG reference channels and subtracts the fitted EOG component, leaving the residual as the cleaned signal. This removes both transient blink artefacts and slow ocular drift without requiring blink event detection.

Usage

correct_eog_regression(
  psg,
  eog_channels = NULL,
  eeg_channels = NULL,
  verbose = TRUE
)

Arguments

Details

This is the pure-R equivalent of MNE's compute_proj_eog() / SSP projection pipeline. Both approaches project out the subspace spanned by the EOG channels; regression does so directly on the continuous signal.

Value

A new mrpheus_psg with cleaned EEG signals and re-segmented epochs. EOG and other channels are unchanged.

Examples

## Not run: 
rec   <- read_edf("psg.edf")
psg   <- prepare_psg(rec) |> preprocess_psg()
clean <- correct_eog_regression(psg)

## End(Not run)

Description

Detects apneas and hypopneas in respiratory airflow and effort signals, returning event-level metadata and summary indices. Detection follows AASM 2012/2017 criteria: >= 90% signal reduction for >= 10 s (apnea), or >= 30% reduction with >= 3% SpO2 desaturation or arousal for >= 10 s (hypopnea).

Usage

detect_apneas(
  psg,
  airflow_channel = NULL,
  spo2_channel = NULL,
  min_duration_s = 10,
  apnea_threshold = 0.9,
  hypopnea_threshold = 0.3,
  desaturation_threshold = 3
)

Arguments

Value

A list with:

events

Tibble. One row per event: epoch, start_s, end_s, duration_s, type ("apnea" / "hypopnea"), desaturation.

summary

Tibble. n_apneas, n_hypopneas, n_events.

See Also

compute_ahi(), compute_odi()

Description

Flags epochs containing likely artefacts based on amplitude thresholds, excessive high-frequency (muscle) power, and movement contamination. Operates on the EEG channels by default. Returns a logical vector (one value per epoch) and optionally attaches per-epoch artefact metrics for inspection.

Usage

detect_artifacts(
  psg,
  channels = NULL,
  amp_threshold_uv = 500,
  hf_band = c(40, 100),
  hf_percentile = 0.99,
  verbose = TRUE
)

Arguments

Value

A tibble with one row per epoch and columns:

epoch

Integer. Epoch index (1-based).

artefact

Logical. TRUE if the epoch is flagged as artefact.

reason

Character. Comma-separated reasons for flagging, or NA.

peak_to_peak_uv

Numeric. Maximum peak-to-peak amplitude across selected channels.

hf_power_db

Numeric. Mean HF band power (dB) across selected channels.

Description

Estimates whether dominant powerline interference is at 50 Hz (Europe/UK) or 60 Hz (North America) by comparing PSD power at both frequencies using Welch's method on the first 10 seconds of the signal.

Usage

detect_powerline(signal, sr)

Arguments

Details

If the signal's Nyquist limit is below 60 Hz, returns 50L immediately.

Value

Integer. 50L or 60L.

Examples

## Not run: 
detect_powerline(eeg_signal, sr = 256)

## End(Not run)

Description

Detects R-peaks in a raw ECG signal using the Pan-Tompkins algorithm. Implements bandpass filtering, derivative, squaring, moving-average integration, and adaptive dual-threshold peak detection with T-wave rejection and search-back for missed beats.

Usage

detect_qrs(ecg, fs)

Arguments

Details

Processing pipeline (Pan & Tompkins, 1985):

1. Bandpass filter 5–15 Hz (Butterworth N=3). At fs = 200 Hz, implemented as separate 12 Hz low-pass then 5 Hz high-pass stages.

2. Derivative filter — 5-point kernel interpolated to match fs.

3. Squaring — non-linear emphasis of dominant peaks.

4. 150 ms moving-average integration.

5. Adaptive thresholding — dual-threshold scheme with T-wave rejection and search-back for missed beats (initialised on 2-second training window).

R-peak indices are returned relative to the bandpass-filtered signal. Account for delay if you need indices aligned to the raw ECG.

Value

A list of class mrpheus_qrs with components:

qrs_i

Integer vector of R-peak sample indices (1-based), referenced to the bandpass-filtered signal.

qrs_amp

Numeric vector of R-wave amplitudes at each detected peak (normalised bandpass signal units).

delay

Numeric. Sample delay introduced by the moving-average stage (half the 150 ms window length).

fs

Numeric. Sampling frequency passed to the function.

References

Pan J, Tompkins WJ. A real-time QRS detection algorithm. IEEE Trans Biomed Eng. 1985;32(3):230–236. doi:10.1109/TBME.1985.325532

See Also

read_philips_physlog(), compute_hr_signal(), compute_hrv_sleep()

Examples

## Not run: 
rec  <- read_philips_physlog("sub-01_physlog.log")
qrs  <- detect_qrs(rec$C[, "v1raw"], fs = rec$HDR$sfreq)
qrs

# Downstream HR signal
hr <- compute_hr_signal(qrs)

## End(Not run)

Description

Prepares the staging tibble produced by stage_epochs() for downstream use with the hypnor package. Attaches recording metadata as attributes and returns a tibble of class mrpheus_hypnogram, which hypnor::new_hypnogram() accepts directly once hypnor is installed.

Usage

export_hypnogram(
  staging,
  epoch_s = 30,
  start_time = NULL,
  participant_id = NULL
)

Arguments

Value

A tibble of class mrpheus_hypnogram with columns epoch, stage, and any probability columns from the staging model. Metadata (epoch_s, start_time, participant_id, source, resolution) are attached as attributes and forwarded to hypnor::new_hypnogram() when hypnor is available.

See Also

stage_epochs()

Description

Designs a 2nd-order Butterworth bandstop filter centred at freq Hz with bandwidth bandwidth_hz and applies it zero-phase using gsignal::filtfilt(). Because the filter has only ~9 coefficients, it is safe to use on long recordings without memory issues.

Usage

notch_filter(signal, sr, freq, bandwidth_hz = 2, harmonics = TRUE)

Arguments

Details

By default also applies the filter at all harmonics of freq below the Nyquist limit, which is the standard approach for powerline noise removal.

Value

Filtered numeric vector, same length as signal.

Examples

## Not run: 
clean <- notch_filter(eeg_signal, sr = 256, freq = 50)
clean <- notch_filter(eeg_signal, sr = 256, freq = 60, harmonics = FALSE)

## End(Not run)

Description

An 8-colour palette extracted from the Roman mosaic Orpheus Charming the Animals (3rd century AD, Palermo Archaeological Museum). The mosaic depicts Orpheus — the mythological figure whose name and story this package honours — surrounded by animals, rendered in warm Mediterranean tesserae.

Usage

palette_orpheus

Format

A named character vector of 8 hex colour codes:

sand

⁠#CDB992⁠ — warm background tessera

vermillion

⁠#B83E2C⁠ — Orpheus's robe; terracotta

olive

⁠#6A7840⁠ — tree and vegetation

umber

⁠#7C5432⁠ — animal fur and earth

bistre

⁠#3C2212⁠ — shadows and outlines

ochre

⁠#B07C3A⁠ — warm amber accent

slate

⁠#6C8284⁠ — birds; dusty teal-grey

ivory

⁠#EAD6AA⁠ — highlights and light tessera

Details

The palette is intentionally earthy and muted, reflecting the natural pigments of Roman mosaic work: sandy limestone backgrounds, terracotta robes, olive vegetation, warm umber fauna, and the distinctive slate-teal of the birds.

Source

Mosaic: Orpheus Charming the Animals, Roman, 3rd century AD. Palermo Archaeological Museum. Image via Wikimedia Commons, https://en.wikipedia.org/wiki/Orpheus.

Examples

if (requireNamespace("scales", quietly = TRUE)) {
  scales::show_col(palette_orpheus)
}

Description

Takes an mrpheus_edf object and segments it into standard epochs, performs a channel inventory (classifying signals by type), and flags channels that appear flat or likely bad. This is the standard entry point before any downstream analyses (spectral, event detection, staging).

Usage

prepare_psg(
  edf,
  epoch_s = 30,
  eeg_pattern = "EEG|C3|C4|F3|F4|O1|O2|Fpz|Pz",
  eog_pattern = "EOG|ROC|LOC",
  emg_pattern = "EMG|chin|Chin",
  ecg_pattern = "ECG|EKG",
  resp_pattern = "Thor|Abdo|Flow|SpO2|airflow",
  flat_threshold = 1e-06
)

Arguments

Value

A list of class mrpheus_psg with components:

edf

The original mrpheus_edf object.

epochs

List. Each element is one epoch (30 s by default), itself a named list of channel vectors.

n_epochs

Integer. Total number of complete epochs.

epoch_s

Numeric. Epoch duration in seconds.

channel_map

Data frame. Channel label, detected type (EEG/EOG/EMG/ECG/RESP/OTHER), sample rate, and bad flag.

Examples

## Not run: 
rec  <- read_edf("data/psg_001.edf")
psg  <- prepare_psg(rec)
psg$channel_map

## End(Not run)

Description

Applies the standard sleep EEG preprocessing pipeline to an mrpheus_psg object: optional channel renaming, DC offset removal, powerline notch filtering (plus harmonics), and channel-type-specific bandpass filtering.

Usage

preprocess_psg(
  psg,
  channel_rename = NULL,
  dc = TRUE,
  powerline_freq = NULL,
  notch_harmonics = TRUE,
  notch_bw_hz = 2,
  eeg_bandpass = c(0.3, 35),
  eog_bandpass = c(0.3, 15),
  emg_bandpass = c(10, 99),
  ecg_bandpass = c(0.5, 40),
  verbose = TRUE
)

Arguments

Details

Filtering is applied to the full continuous signal stored in psg$edf$signals before re-epoching, which avoids filter discontinuities at epoch boundaries. The returned object is a new mrpheus_psg with the same epoch length and channel map as the input.

The individual filter steps (remove_dc(), detect_powerline(), notch_filter(), bandpass_filter()) are exported separately and can be called on any numeric vector without constructing a PSG object.

Value

A new mrpheus_psg object with filtered signals and re-segmented epochs. The edf$signals entries are replaced in-place with the filtered continuous signals so that subsequent epoch access is consistent.

Examples

## Not run: 
rec  <- read_edf("data/psg_001.edf")
psg  <- prepare_psg(rec)

# Auto-detect powerline, apply standard filters
psg_clean <- preprocess_psg(psg)

# Rename linked-ear channels to 10-20 names, fix powerline explicitly
psg_clean <- preprocess_psg(
  psg,
  channel_rename = c("C3-A2" = "C3", "C4-A1" = "C4",
                     "O1-A2" = "O1", "O2-A1" = "O2"),
  powerline_freq = 50L
)

## End(Not run)

Description

Reads a European Data Format (EDF or EDF+) file and returns a structured list containing signal data, channel metadata, and recording header. Wraps edfReader::readEdfHeader() and edfReader::readEdfSignals() with consistent output formatting for the mrpheus pipeline.

Usage

read_edf(path, channels = NULL, only_header = FALSE)

Arguments

Value

A list of class mrpheus_edf with components:

header

Data frame. Recording metadata (patient info, start time, number of signals, etc.).

signals

Named list of numeric vectors, one per channel.

channels

Data frame. Channel-level metadata: label, sample rate, physical min/max, digital min/max, transducer type, prefiltering.

duration_s

Numeric. Total recording duration in seconds.

path

Character. Resolved path to the source file.

Examples

## Not run: 
rec <- read_edf("data/psg_001.edf")
rec <- read_edf("data/psg_001.edf", channels = c("EEG Fpz-Cz", "EOG horizontal"))
rec$channels

## End(Not run)

Description

Reads a Philips PMU physiological log (.log) file recorded alongside an MRI acquisition and returns a structured object containing the signal matrix, event markers, and recording metadata. Supports all Philips PMU hardware variants via the system preset.

Usage

read_philips_physlog(
  path,
  system = c("wBTU", "wired", "custom"),
  sfreq = NULL,
  channels = NULL,
  skipprep = FALSE
)

Arguments

Details

Bit-encoded marker values:

0x0001  ECG trigger       0x0002  PPU trigger      0x0004  respiration
0x0008  slice onset       0x0010  scanner start    0x0020  scanner stop
0x0040  external trigger  0x0080  calibration      0x8000  ref ECG trigger

Use the scanner-stop marker (more reliable than start) to time-lock data:

stopidx  <- tail(data$I$ScannerStop, 1) - enddelay
startidx <- stopidx - round(n_vols * TR * data$HDR$sfreq)
epoch    <- data$C[startidx:stopidx, ]

Value

A list of class mrpheus_physlog with components:

C

Integer matrix (samples × channels). Column order matches the file header.

M

Two-column integer matrix: marker (bit-encoded value) and index (1-based sample number) for every non-zero marker.

I

Named list of 1-based sample index vectors for each event type: VcgOnset, PpuOnset, TriggerResp, Measurement, ScannerStart, ScannerStop, TriggerExt, Calibration, RefTriggerVcg.

HDR

List of header metadata: ID, DATETIME, STATS, DockableTable, COLUMN_NAMES, system, sfreq.

path

Character. Resolved path to the source file.

See Also

detect_qrs(), compute_hr_signal()

Examples

## Not run: 
# Philips Achieva/Ingenia with wireless VCG (default)
rec <- read_philips_physlog("sub-01_ses-01_physlog.log")
rec

# Older wired scanner
rec <- read_philips_physlog("sub-01_physlog.log", system = "wired")

# Custom sampling rate
rec <- read_philips_physlog("sub-01_physlog.log", system = "custom", sfreq = 400)

# ECG channels only
rec <- read_philips_physlog("sub-01_physlog.log", channels = c("v1raw", "v2raw"))

## End(Not run)

Description

Subtracts the signal mean (per-channel DC offset removal). Should be applied before filtering to prevent filter transients from a non-zero baseline.

Usage

remove_dc(signal)

Arguments

Value

Numeric vector with mean removed, same length as signal.

Examples

x <- c(100.1, 100.5, 99.8, 100.3)
remove_dc(x)

Description

Stages each 30-second epoch using a pre-trained LightGBM model originally developed for YASA (Vallat & Walker, 2021) and shipped as a cross-language serialised model in inst/models/yasa_staging.txt. Features are computed in R to match the Python feature extraction pipeline exactly; bit-exact parity is validated in the package test suite.

Usage

stage_epochs(
  psg,
  eeg_channel = NULL,
  eog_channel = NULL,
  emg_channel = NULL,
  artefacts = NULL,
  model_path = system.file("models", "yasa_staging.txt", package = "mrpheus")
)

Arguments

Details

Stages returned follow standard AASM nomenclature: W (wake), N1, N2, N3, REM.

Value

A tibble with one row per epoch:

epoch

Integer.

stage

Character. AASM stage: W, N1, N2, N3, REM, or NA (artefact).

prob_W, prob_N1, prob_N2, prob_N3, prob_REM

Numeric. Posterior class probabilities from the LightGBM model.

References

Vallat, R., & Walker, M. P. (2021). An open-source, high-performance tool for automated sleep staging. eLife, 10, e70092. doi:10.7554/eLife.70092

See Also

export_hypnogram()