fastEIT - Parser Reference

Specifications for all parsers: input formats, output containers, calibration details, and the pyEIT reconstruction bridge.

For data container field specifications see data_model.md. For architecture and extension recipes see parsing_layer.md.

---

1. Overview

File	Vendor	Parser	Container
.bin	Dräger	DragerBinParser	ReconstructedFrameData
.asc / .txt / .csv	Dräger	DragerAscParser	ContinuousSignalData
.eit	Dräger	DragerEitParser	RawImpedanceData
.csv / .txt / .asc	Timpel	TimpelTabularParser	ReconstructedFrameData

All parsers inherit from BaseParser and are accessible via load_data(path).

---

2. DragerBinParser - .bin

File: parsers/draeger/bin/bin_parser.py

Reads vendor-reconstructed 32×32 pixel frames exported by the PulmoVista 500. Two frame formats are registered; detection is size-based (no header).

Frame formats

Name	Frame size	Medibus channels	Notes
Draeger_base_4358	4358 bytes	52	Standard export
Draeger_ext_4382	4382 bytes	58	BiLevel parameters + 4 PressurePod channels; the channels are there whether or not a Pod is connected, metadata["has_pressure_pod"] says if they hold values

Binary frame layout (base format)

Field	Type	Size	Content
ts	float64	8 B	Wall-clock timestamp as fraction of day (kept as-is; × 86400 gives seconds)
dummy	float32	4 B	Unknown - ignored
pixels	float32 × 1024	4096 B	Reconstructed 32×32 image (C-order)
min_max_flag	int32	4 B	+1 = inspiratory peak, −1 = expiratory trough, 0 = none
event_marker	int32	4 B	Event counter
event_text	S30	30 B	ASCII event label, padded with spaces (or nulls on frames without an event)
timing_error	int32	4 B	0 = frame timing OK
medibus_data	float32 × 52	208 B	Ventilator Medibus channels; sentinel −3.4×10³⁸ = disconnected

The two formats share Medibus indices 0-50 and differ from index 51 onward:

Index	Base (52 channels)	PressurePod (58 channels)
51	time_at_low_pressure	high_pressure (BiLevel PHigh)
52	-	low_pressure (BiLevel Plow)
53	-	time_at_low_pressure (BiLevel Tlow)
54-57	-	Pod waveforms: airway_pressure_pod, esophageal_pressure_pod, transpulmonary_pressure_pod, gastric_pressure_pod

Full field list with units: parsers/draeger/bin/draeger_dtypes.py.

Sentinel values

Disconnected Medibus channels carry the sentinel −3.4×10³⁸ (IEEE 754 bit pattern 0xFF7FC99E); BiLevel fields (51-53 in the PressurePod format) read −1000.0 when the ventilator is not in BiLevel mode. The parser returns these values unchanged. Replacing them with NaN is a preprocessing step; bin_utils.replace_no_data_sentinels() is available for it.

Sampling frequency

Not stored in the binary. Estimated from the timestamps: fs = (N − 1) / (t_last − t_first), with timestamps converted to seconds and unwrapped across midnight, rounded to the nearest integer Hz. Falls back to 50 Hz (recorded in metadata["warnings"]) when the estimate fails.

---

3. DragerEitParser - .eit

File: parsers/draeger/eit/eit_parser.py

Reads raw transimpedance measurements from the PulmoVista 500 proprietary binary format. We are not aware of another open-source Python parser for it.

File structure

Bytes 0-11          : preamble - 3 × int32 LE
                        [0:4]  format_version (always 51)
                        [4:8]  sep_offset      (file-specific, variable)
                        [8:12] unknown_int
Bytes 12-sep_offset : ASCII header ("key: value\r\n" lines, latin-1)
Bytes sep_offset-+7 : separator b'**\r\n\r\n\r\n' (8 bytes)
Bytes sep_offset+8… : binary frames, 5495 bytes each

ASCII header fields

Header key	Metadata key	Type	Notes
Framerate [Hz]	fs	float	Acquisition rate (typically 50.0 Hz)
Date	date	str	DD.MM.YYYY
Time	time	str	HH:MM:SS.mmm
Frequency [kHz]	frequency_khz	float	Excitation frequency (≈101.5 kHz)
Amplitude [uA]	amplitude_ua	float	Injection current amplitude (≈9100 µA)
Gain	gain	int	Hardware amplifier gain
Samples	samples_per_period	int	ADC samples per excitation period
Periods	periods	int	Excitation periods per measurement

Binary frame layout (5495 bytes)

Field	Type	Size	Content
timestamp	float64	8 B	Fraction of day
unknown_f8	float64	8 B	Unknown
trans_A	float64 × 208	1664 B	Primary transimpedance set (ADC counts)
unknown_16a	float64 × 16	128 B	Unknown
injection_current	float64 × 16	128 B	Injected current per drive pattern (ADC counts)
unknown_16b	float64 × 16	128 B	Unknown
voltage_A	float64 × 16	128 B	Electrode voltage A (ADC counts)
unknown_50	float64 × 50	400 B	Unknown
trans_B	float64 × 208	1664 B	Reference transimpedance set (ADC counts)
unknown_48	float64 × 48	384 B	Unknown
voltage_B	float64 × 16	128 B	Electrode voltage B (ADC counts)
unknown_6	float64 × 6	48 B	Unknown
gugus	float64 × 44	352 B	Ignored (EIDORS terminology)
unknown_byte	uint8	1 B	Unknown
medibus	float32 × 67	268 B	Ventilator Medibus channels (67 in .eit, 52 in .bin)
event_text	S30	30 B	ASCII event label
mixed	uint8 × 24	24 B	Mixed content
frame_counter	uint16	2 B	Rolling frame counter
padding	uint8 × 2	2 B	Alignment padding

Calibration pipeline

Empirical constants from EIDORS read_draeger_file.m (A. Adler, estimated 2016-04-07). These are not present in the .eit header - they are hardware-specific and hardcoded.

Constant	Value	Role
FT_A	0.00098242	Transimpedance scaling - primary channel
FT_B	0.00019607	Transimpedance scaling - reference channel
FC_CURRENT	194326.3536	ADC counts → Ampere
FV_VOLTAGE	0.11771	ADC counts → Volt

vv      = FT_A * trans_A - FT_B * trans_B   # (N_frames, 208) calibrated transimpedance [Ω]
I_real  = injection_current / FC_CURRENT     # (N_frames, 16)  injected current [A]
V_diff  = (voltage_A - voltage_B) / FV_VOLTAGE  # (N_frames, 16) differential voltage [V]

Measurement protocol

208 measurements per frame = 16 electrodes × 13 measurement pairs. Adjacent current injection (drive distance = 1 electrode), adjacent voltage measurement (step = 1 electrode).

---

4. DragerAscParser - .asc

File: parsers/draeger/asc/asc_parser.py

Reads the PulmoVista 500 continuous waveform export (tab-separated text). Returns ContinuousSignalData with one row per EIT frame.

Detection: looks for the DraegerEIT Software header string in the first 40 lines (read as latin-1). Sampling frequency is estimated from the time column (fraction of day) when present, otherwise fs is None.

Typical columns: image, time, global, local_1_x_16_y_29 … local_4_x_16_y_05, minmax, event, eventtext, timing_error, plus Medibus channel columns when the ventilator is connected via Medibus.

---

5. TimpelTabularParser - .csv / .txt / .asc

File: parsers/timpel/timpel_parser.py

Reads Timpel EIT device tabular exports. Each row is one EIT frame; columns 0-1023 are the 32×32 pixel values (C-order); columns 1024-1029 are auxiliary signals (airway_pressure, flow, volume, min_flag, max_flag, qrs_flag).

Returns ReconstructedFrameData. The export has no timestamp column, so fs is fixed at 50 Hz and ts is generated as (frame_index + first_frame) / 50.

Detection: the keyword timpel in the first 40 lines, or a first data row with at least 1029 separators (comma, tab or semicolon), i.e. the 1030-column layout.

---

6. GREIT reconstruction bridge

This function is far from reliable. It was written only to show one possible application of the .eit parser output: passing it to a reconstruction library such as pyEIT.

File: parsers/draeger/eit/eit_pyeit_bridge.py Optional dependency: pip install fasteit[pyeit]

Converts RawImpedanceData.measurements (N_frames, 208) to reconstructed 32×32 images (N_frames, 32, 32) using the GREIT algorithm.

Reference: Adler A et al., "GREIT: a unified approach to 2D linear EIT reconstruction of lung images." Physiol. Meas. 30 (2009) S35-S55. DOI: 10.1088/0967-3334/30/6/S03

build_greit(n_el, h0, p, lamb, n) → (solver, protocol)

Builds a pyEIT GREIT solver on a circular unit-disk mesh.

Parameter	Default	Meaning
n_el	16	Number of electrodes
h0	0.1	Mesh density
p	0.2	GREIT noise figure (Adler 2009 recommendation)
lamb	1e-2	Tikhonov regularisation
n	32	Output image side length in pixels

reconstruct_greit(measurements, ref_frame, …) → ndarray

ref_frame value	Behaviour
None	Mean of all frames in this recording (default)
int	Single frame index
(start, end)	Mean of measurements[start:end]
np.ndarray (208,)	External reference - for cross-recording EELI comparison

Sign convention: GREIT outputs conductivity change (Δσ); the function negates it to match the Dräger impedance convention (air → impedance rises → positive peak).

Image orientation - three steps applied inside reconstruct_greit(), all sourced from PulmoVista 500 IFU SW 1.3n: 1. Negate (sign flip): impedance convention - see above. 2. rot90(k=1) (90° CCW): electrode 1 sits at 12 o'clock on the PulmoVista belt (anterior/ventral wall); pyEIT places electrode 0 at 3 o'clock. Rotation moves the anterior region to the top of the image. Source: IFU p. 144. 3. fliplr (horizontal mirror): Dräger displays images in caudal-cranial projection (CT convention viewed from feet): left side of image = right side of patient. pyEIT produces the anatomical mirror; fliplr corrects this. Source: IFU p. 146.

Spatial quality note: a generic circular mesh does not replicate Dräger's proprietary reconstruction. Both algorithms produce correct relative distributions, but the vendor .bin uses a device-specific mesh and matrix.

---

7. Loader utilities

File: parsers/loader.py

load_data(path) → BaseData

Single entry point. Auto-detects vendor and format, selects the right parser, returns the appropriate container. Raises ValueError for an unsupported extension or unrecognised content, NotImplementedError when no parser is registered for the detected vendor/extension, and a ParserError subclass for .bin files with an unknown or ambiguous frame size.

from fasteit.parsers.loader import load_data

data = load_data("patient01.bin")   # → ReconstructedFrameData
data = load_data("patient01.eit")   # → RawImpedanceData
data = load_data("patient01.asc")   # → ContinuousSignalData

load_many(paths) → list[BaseData]

Parses an explicit list of paths in order. Formats can be mixed.

from fasteit.parsers.loader import load_many

recordings = load_many(["patient01.bin", "patient02.bin", "patient01.asc"])

load_folder(folder, pattern="*") → list[BaseData]

Scans a directory, parses every file with a registered parser. Files that raise an error are skipped and a one-line message is printed to stderr.

from fasteit.parsers.loader import load_folder

all_bin  = load_folder("/data/session01/", pattern="*.bin")
all_data = load_folder("/data/session01/")          # mixed formats
recursive = load_folder("/data/", pattern="**/*.bin")