wfdb

A PhysioNet WFDB toolbox for Nelson.

WFDB is the record format the PhysioNet databases are published in: a text header (.hea) that describes the signals, a binary signal file (.dat) that packs their samples, and optional MIT annotation files (.atr and the like) that mark the beats and events.

Load

run('/path/to/wfdb/loader.m')

Functions

All functions live in the wfdb package (call them as wfdb.read, wfdb.write, ...).

Function	Purpose
wfdb.read(name, ...)	Read a record (header, signals, annotations) into the neutral struct. Options: 'Physical', 'Signals', 'Annotations', 'Data'.
wfdb.write(name, r, ...) / wfdb.write(name, samples, rate, ...)	Write the neutral struct, or a plain matrix. Options: 'Format', 'Physical', 'Gain', 'Baseline', 'Leads', 'Units', 'Annotations', 'Segments'.
wfdb.header(name)	Read only the header, no samples and no annotations, fast.
wfdb.info(name)	A compact human-readable summary (header only).
wfdb.iswfdb(name)	Probe a record; true when its header parses, never throws.
wfdb.annotations(file, ...)	Read a MIT annotation file, or write one back.
wfdb.validate(name)	Check a record: header, signal-file length and the stored checksums.
wfdb.matrix(r)	Turn the per-signal struct into one aligned matrix, a column per signal.
wfdb.select(r, which)	Keep a subset of signals, by label or index, in the order asked.
wfdb.crop(r, [t0 t1], ...)	Keep a time (or 'Frames') window; the annotations in it come along.

wfdb.read options: 'Physical', true returns physical units ((raw - baseline) / gain) instead of the raw ADC integers the file holds (raw is the default, so a read followed by wfdb.write is lossless, with the gain, baseline and unit of every signal on the struct); 'Signals', which keeps a subset by label or index; 'Annotations' chooses the annotation file ('auto' reads name.atr when present and skips it when absent, an extension such as 'qrs' reads name.qrs, a path reads that file, false reads none); and 'Data', false reads the header only.

Signal formats read: 8, 16, 24, 32, 61, 80, 160, 212, 310 and 311. Every one of these except the first-difference format 8 is also written. MIT annotations round-trip in full: sample, time, code, symbol, num, subtype, chan and aux.

The neutral struct

wfdb.read returns one struct with no domain assumptions:

r.kind            % 'wfdb.record'
r.name, r.path
r.rate            % the base sampling frequency (Hz)
r.counterFrequency, r.baseCounter, r.declaredFrames
r.startTime, r.startTimeKnown, r.startDate, r.comments
r.multiSegment, r.multiFrequency, r.segments
r.signals(k)      % .file .format .spf .offset .skew .gain .baseline .unit
                  % .adcres .adczero .initial .checksum .blocksize
                  % .description .label .rate .data
r.annotations     % .sample .time .code .symbol .num .subtype .chan .aux

wfdb.write accepts the same shape, or a plain matrix (one column per signal) with a rate; the gains, baselines, labels and units are used when present and defaulted when not. With 'Segments', lengths it writes a fixed-layout multi-segment record (the signals cut into consecutive pieces, each an ordinary record, named by a master header); lengths is a vector of per-segment frame counts or a scalar chunk size, and wfdb.read stitches the pieces back into the one record that went in.

Example

A ready-to-run tour is in examples/read_write_demo.m.

% Round-trip a two-lead record and read it back in physical units.
rate = 250;
t = (0:5 * rate - 1)' / rate;
wfdb.write('demo', [sin(2*pi*1.3*t), 0.5*cos(2*pi*1.3*t)], rate, ...
  'Leads', {'I', 'II'}, 'Units', {'mV', 'mV'});
r = wfdb.read('demo', 'Physical', true);
m = wfdb.matrix(r);

Dependencies

This module depends on no other Nelson module.

License

LGPL-3.0-or-later.