fastEIT - Clinical Indices (layer 3)¶

Layer 3 turns a preprocessed recording (EITRecording, preprocessing.md) or a segmented PEEP trial (PeepTrial, segmentation.md) into tables of clinical indices: one row per breath, per breath and region, per PEEP step, per step and region, and a summary. The tables are ready for a database or for R: the clinical context (patient, position, side in park bench) is joined at the time of the analysis, the recording does not need to know it.

The definitions and the evidence are in notebooks/07_clinical_indices.ipynb (11 tidal recordings, four PEEP trials); on the four trials the functions below give the crossing PEEP, the best PEEP and the change of EELI in mL of the notebook.

Steps for a user: analyse a recording.

Entry points
Indices of a breath
Tidal recordings
PEEP trials
Limits

1. Entry points¶

from fasteit.features import peep_trial_features, tidal_features
from fasteit.models import RecordingContext, Segment
from fasteit.parsers.loader import load_data
from fasteit.preprocessing import preprocess
from fasteit.segmentation import segment_peep_trial

# A recording at one setting (or the tidal phase of a longer one).
recording = preprocess(load_data("patient01.bin"),
                       context=RecordingContext(tidal_volume_ml=450))  # Vt optional
tidal = tidal_features(recording)            # or phase=Segment(0, 300, category="phase", label="tidal")
tidal.breaths    # one row per breath
tidal.regions    # one row per breath and region
tidal.summary    # median and quartiles over the valid breaths, by region and index

# A PEEP trial.
context = RecordingContext(protocol="peep_trial", tidal_volume_ml=450)
trial = segment_peep_trial(preprocess(load_data("peep_trial.bin"), context=context))
features = peep_trial_features(trial)   # parameters in features.config (AnalysisConfig)
features.steps          # one row per PEEP step
features.step_regions   # one row per step and region
features.summary        # crossing PEEP, best PEEP, R/I (global and by region)

The object holds everything: features.trial.steps (driving pressure and tidal volume of each step, breaths analysed), features.trial.breaths (step of each breath), features.trial.recording.breaths (times, quality flags, PEEP, plateau, driving pressure, measured tidal volume). Each value is stored once: the tables of layer 3 have the keys (breath_id, step, peep, region) and what layer 3 computes, and are joined on the keys with the tables before them. The flat tables with everything in one row per breath and per step, for R, SPSS and databases, are made by layer 4 (export.md).

2. Indices of a breath¶

All from the frames of the layer 2 breath (one sample, no averaging): start = end-expiratory valley, peak = end-inspiratory peak.

Column	Definition	Source
`eeli`	signal inside the lung mask at the start of the breath (arbitrary units)	TREND, Frerichs 2017, DOI: 10.1136/thoraxjnl-2016-208357
`tiv`	signal at the peak minus at the start	TREND
`gi`	global inhomogeneity: sum of \|pixel TIV - median\| / sum of pixel TIV, over the lung pixels	Zhao 2009, DOI: 10.1007/s00134-009-1589-y
`cov_vd`, `cov_rl`	centre of ventilation on the image grid: 0 = ventral / patient right, 1 = dorsal / patient left	Frerichs 2006, DOI: 10.1164/rccm.200512-1942OC
`respiratory_rate`	60 / duration of the EIT breath (breaths/min)	-
`tidal_volume_used`, `tidal_volume_source`	tidal volume used for the values in mL: typed for the step, typed for the recording, or measured	layer 2 and 2b
`compliance`	TIV / driving pressure (EIT units per cmH2O)	-
`compliance_ml`	tidal volume / driving pressure (mL/cmH2O): static compliance of the respiratory system, with the tidal volume typed or measured	-

By region (regions, one row per breath and region; regions of layer 2: ventral and dorsal halves, layers, right and left, quadrants):

Column	Definition
`eeli`, `tiv`	the same on the signal of the region
`share`	TIV of the region / TIV of the lung (0-1): a ratio of sums
`tiv_ml`	share × tidal volume (mL)
`compliance`	TIV of the region / driving pressure (EIT units per cmH2O): the sum of the compliances of its pixels, which are in parallel under the same driving pressure
`compliance_ml`	share × tidal volume / driving pressure (mL/cmH2O)

The two compliances are both kept: the one in mL/cmH2O is the static compliance of the respiratory system; the one in EIT units also follows what changes in the imaged slice (in volume control the same tidal volume gives 80-105 % of the TIV of the first step across a PEEP trial) and can move the best PEEP (B park bench: 4.5 in EIT units, 8.5 in mL/cmH2O). Costa is computed in EIT units only. A region's compliance is a partial compliance, its share of the total: a small region has a low compliance even when it is healthy.

3. Tidal recordings¶

tidal_features(recording, phase=None) gives the indices of every breath, and a summary over the valid breaths (no quality flag, not excluded) inside the tidal phase (a breath is inside when it starts and ends inside it): one row per region ("global" for the whole lung) and index (feature), with median, q1, q3 and n_breaths (breaths with a value). The tidal volume of the breaths is the one typed in RecordingContext.tidal_volume_ml, otherwise the one measured on the ventilator curves.

4. PEEP trials¶

peep_trial_features(trial) gives the indices of every breath (with peep_step and analysed) and, for each step, from its analysed breaths:

Column	Definition
`respiratory_rate`, `eeli`, `tiv`, `gi`, `cov_vd`, `cov_rl`	median of the analysed breaths
`compliance`	median TIV / driving pressure of the step: it also works when the driving pressure comes from one inspiratory hold or is typed
`compliance_ml`	tidal volume / driving pressure of the step (it needs no EIT data, so it exists also for a step without analysed breaths)
`compliance_pct_of_best`, `compliance_ml_pct_of_best`	as % of the best step
`collapse`, `overdistension`	Costa method (%), from the median tidal image of the step: pixel compliance, loss as % of the best of each pixel, weighted by the best compliance; the loss of a pixel is at most 100 % (Costa 2009, DOI: 10.1007/s00134-009-1447-y; formulas as in Shono and Kotani 2019, DOI: 10.1186/s40560-019-0358-4)
`d_eeli_tiv_vs_previous`, `d_eeli_tiv_vs_first`	change of EELI in tidal volumes: change / TIV of the step at the lower PEEP of the two (Wang 2020, DOI: 10.12659/MSM.922609)
`d_eeli_ml_vs_previous`, `d_eeli_ml_vs_first`	the same in mL: × tidal volume of that step
`d_compliance_pct_of_best_vs_...`, `d_compliance_ml_pct_of_best_vs_...`	changes of compliance vs the previous and the first step, in points of % of the best

Collapse and overdistension are read on their curves (one value per step); a change between two PEEP levels is the difference of two values of the curve.

step_regions has, for each step and region, the medians of eeli, tiv and share of the analysed breaths, and tiv_ml, compliance and compliance_ml of the step.

The summary is a long table (region, feature, value):

Feature	Region	Definition
`crossing_peep`	global	PEEP where collapse and overdistension cross (linear interpolation; the lowest if they cross more than once)
`best_peep_compliance`, `best_peep_compliance_ml`	global	PEEP of the best global compliance, in EIT units and in mL/cmH2O
`ri_peep_high`, `ri_peep_low`	global	PEEP of the two steps used for R/I
`ri`	global and each region	recruitment-to-inflation ratio
`recruited_volume_ml`	global and each region	recruited volume (mL, with a tidal volume)

The PEEP of crossing_peep, the best PEEP and the R/I steps are on the levels of the protocol (peep_set, from SegmentationConfig.peep_levels) when every step has one, so that they can be compared between patients; otherwise on the measured PEEP. The R/I steps are chosen, and the drop of PEEP in its formula is computed, on the measured PEEP (the change of pressure that happened).

Recruitment-to-inflation ratio (Chen et al., Am J Respir Crit Care Med 2020, DOI: 10.1164/rccm.201902-0334OC), between the step nearest to PEEP 15 and the step nearest to PEEP 5 at least 10 below it (AnalysisConfig):

dEELV = (EELI_high - EELI_low) x Vt_low / TIV_low        (Wang 2020)
recruited volume = dEELV - C_low x dPEEP                 C_low = Vt_low / driving_low
R/I = (recruited volume / dPEEP) / C_low = (EELI_high - EELI_low) / (C_EIT,low x dPEEP) - 1

The tidal volume cancels in R/I, which needs only EIT and the driving pressure; a region uses its own EELI and compliance. The original method reads one expiration after the drop of PEEP; here each level is the equilibrium EELI of a step, as in Wang 2020.

Steps with fewer analysed breaths than the minimum of layer 2b (enough false) are kept, with a warning: filter them in the analysis if needed. If the first step has no analysed breaths, the changes vs the first step are NaN.

5. Limits¶

Tested on four decremental PEEP trials from two patients and on synthetic data with a known answer; Costa is implemented from its published formulas, not checked against another implementation.
EELI is in arbitrary units: only its changes count. Its change in mL assumes the same impedance per mL for the tidal and the end-expiratory change (Hinz et al., Intensive Care Med 2003, DOI: 10.1007/s00134-002-1555-4); in the four trials it was 1.6-2.3 times compliance × change of PEEP, cause unknown, so it is exploratory.
The mL of a region share the tidal volume as seen in the imaged slice: a distribution, not independent measurements.
Compliance is that of the respiratory system, in a passive patient; regional compliance uses the same driving pressure for every region.
% of the best step, Costa and the changes vs the first step depend on the PEEP range tested: compare patients with the same steps.
Costa reads an error common to a step (its driving pressure, the share of the tidal volume seen by the slice) as collapse or overdistension.
R/I mixes recruitment and overdistension (Zhao et al., Crit Care 2023, DOI: 10.1186/s13054-023-04520-8) and becomes large in a region almost collapsed at the low PEEP.
One lung mask for the whole recording: GI is not comparable with cut-offs computed on other masks.

Research use only - not a medical device.