Respiratory (RSP)

Respiratory rate is often measured on the chest using a respiration belt or a respiratory inductance plethysmography (RIP) sensor. physioKIT provides a set of functions to process RSP signals. The functions can be used to generate synthetic RSP signals, clean noisy RSP signals, extract respiratory peaks, compute respiratory rate, and compute dual band metrics.

The snippets below form a walkthrough: later snippets may reuse variables from earlier ones. The plots are saved example outputs. For a standalone runnable example, start with the quickstart.

Synthetic RSP

We can generate a synthetic RSP signal using the rsp.synthesize function. The function returns three NumPy arrays: the RSP waveform, segmentation mask, and fiducial mask. The signal_length parameter specifies the number of samples. The sample_rate parameter specifies the sampling rate in Hz. The respiratory_rate parameter specifies the respiratory rate in breaths per minute (BPM).

In the following snippet, we generate a synthetic RSP inductance band signal with a respiratory rate of 12 BPM sampled at 1000 Hz.

import physiokit as pk

sample_rate = 1000 # Hz
respiratory_rate = 12 # BPM
signal_length = 10 * sample_rate # 10 seconds

rsp, segs, fids = pk.rsp.synthesize(
    signal_length=signal_length,
    sample_rate=sample_rate,
    respiratory_rate=respiratory_rate,
)

---

Noise Injection

We can additionally add noise to generate a more realistic RSP signal.

Below adds baseline wander, powerline noise, and custom noise sources to the synthetic RSP signal.

# Add baseline wander
rsp_noise = pk.signal.add_baseline_wander(
    data=rsp,
    amplitude=2,
    frequency=.05,
    sample_rate=sample_rate
)

# Add powerline noise
rsp_noise = pk.signal.add_powerline_noise(
    data=rsp_noise,
    amplitude=0.05,
    frequency=60,
    sample_rate=sample_rate
)

# Add additional noise sources
rsp_noise = pk.signal.add_noise_sources(
    data=rsp_noise,
    amplitudes=[0.05, 0.05],
    frequencies=[10, 20],
    noise_shapes=["laplace", "laplace"],
    sample_rate=sample_rate
)

---

Sanitize RSP

We can clean the RSP signal using the rsp.clean function. By default, the routine implements a bandpass filter with cutoff frequencies of 0.05 Hz and 3 Hz. The lowcut and highcut parameters can be used to specify the cutoff frequencies. The order parameter specifies the order of the filter. The sample_rate parameter specifies the sampling rate in Hz. The function returns a numpy array with the cleaned RSP signal.

In the following snippet, we clean the noisy RSP signal using a bandpass filter with cutoff frequencies of 0.05 Hz and 3 Hz.

...

# Clean RSP signal
rsp_clean = pk.rsp.clean(
    data=rsp_noise,
    lowcut=0.05,
    highcut=3,
    order=3,
    sample_rate=sample_rate
)

Extract Respiratory Peaks

A common task in RSP processing is to extract respiratory peaks. The rsp.find_peaks function implements a simple peak detection algorithm. The function returns a numpy array with the indices of the peaks.

In the following snippet, we extract respiratory peaks from the cleaned RSP signal.

...

# Extract respiratory cycles
peaks = pk.rsp.find_peaks(data=rsp_clean, sample_rate=sample_rate)

# Compute RR-intervals
rri = pk.rsp.compute_rr_intervals(peaks=peaks)

# Filter RR-intervals
mask = pk.rsp.filter_rr_intervals(rr_ints=rri, sample_rate=sample_rate)

# Keep normal RR-intervals
peaks_clean = peaks[mask == 0]
rri_clean = rri[mask == 0]

---

Compute Respiratory Rate

The rsp.compute_respiratory_rate function computes the respiratory rate in breaths per minute (BPM) based on the selected method. The peak method computes respiratory rate based on identified respiratory peaks whereas fft method uses FFT to compute the respiratory rate. The function returns the respiratory rate in breaths per minute (BPM) along with a “quality of signal” (QoS) metric. The QoS metric is based on the selected method.

In the following snippet, we compute the respiratory rate from the cleaned waveform using the FFT method.

# Compute respiratory rate
rr_bpm, rr_qos = pk.rsp.compute_respiratory_rate(
    data=rsp_clean,
    method="fft",
    sample_rate=sample_rate,
    lowcut=0.05,
    highcut=1
)

Example output:
Respiratory rate: 12 BPM

---

Compute Dual Band Metrics

Using dual RIP bands, a ribcage (RC) band and an abdominal (AB) band, we can compute additional respiratory metrics. The rsp.compute_dual_band_metrics function computes the following metrics:

Metric	Description
rc_rr	RC respiratory rate (BPM)
ab_rr	AB respiratory rate (BPM)
vt_rr	VT respiratory rate (BPM)
phase	Phase angle (degrees)
lbi	Labored breathing index
rc_lead	RC leads AB
rc_percent	Percent RC contribution
qos	Quality of signal (0-1)
rc_pk_freq	RC peak frequency (Hz)
rc_pk_pwr	RC peak power
ab_pk_freq	AB peak frequency (Hz)
ab_pk_pwr	AB peak power
vt_pk_freq	VT peak frequency (Hz)
vt_pk_pwr	VT peak power

In the following example, we generate synthetic RC and AB band data, compute the dual band metrics, and plot the results. We compute the metrics over a sliding window with a length of 10 seconds and an overlap of 1 second.

sample_rate = 1000
respiratory_rate = 12.2
rc_amp = 1.5
ab_amp = 1.0
dur_sec = 60
win_len = 10*sample_rate
ovl_len = 1*sample_rate

# Synthesize RC and AB band data
rc, segs, fids = pk.rsp.synthesize(
    signal_length=dur_sec*sample_rate,
    sample_rate=sample_rate,
    respiratory_rate=respiratory_rate
)
rc = rc * rc_amp

ab, segs, fids = pk.rsp.synthesize(
    signal_length=dur_sec*sample_rate,
    sample_rate=sample_rate,
    respiratory_rate=respiratory_rate
)
ab = ab * ab_amp

# Compute metrics over sliding window
ts_metrics, dual_metrics = [], []
for i in range(0, rc.size - win_len, ovl_len):
    rc_win = rc[i:i+win_len]
    ab_win = ab[i:i+win_len]
    ts_metrics.append((i+win_len/2)/sample_rate)
    dual_metrics.append(pk.rsp.compute_dual_band_metrics(
        rc=rc_win,
        ab=ab_win,
        sample_rate=sample_rate,
        pwr_threshold=0.9
    ))

---

API

Refer to RSP API for more details