In-vitro dataset for classification and regression of stenosis: dependence on heart rate, waveform and location

Background

This data supplements the paper "Classification and regression of stenosis using an in-vitro pulse wave dataset:
dependence on heart rate, waveform and location".  It was created at Technische Hochschule Mittelhessen (THM) in Germany and uploaded to Zenodo. Please cite the paper and the Zenodo doi when using this dataset.

General description / Dataset structure

Each mat-File describes a different measurement (details can be found in the paper). There are 17 pressure signals for different positions, one flow sensor close to the stenosis location and one monitor signal of the proportional valve use to control the input curve. Total duration of each signal is 60s with a sampling rate of 1000 Hz. Each mat-file contains a header structure with metadata and struct array for signals of each sensor. Signals in each mat-File are aligned with respect to a common time axis, but this is not guaranteed between different measurements/files. We did our best to make the beginnings end endings align as close as possible (by removing buffer artefacts and aligning the input signal of the monitor), however algorithms should not rely on a global time axis. This similar to patient measurements without an ekg, this does also not share a global time axis comparable among patients.

The file format can either be loaded directly in Matlab or in Python with scipy's loadmat function.

The data is structure first by stenosis "state" (or location) then by heart rate and then by heart waveform. The stenosis "states" can devided in 1 subset of 10 folders created for regression and 6 created for classification. Excerpt of the folder structure:

No Stenosis

HR 50

WaveForm1.mat

WaveForm2.mat

...

HR 55

...

...

Regression - Stenosis at Pos01

HR 50

...

...

...

The tools also available at this page help with traversing this folder structure and are available for Python and Matlab.

Data Fields of each file

headerStruct

field
description

id
internal database id

name
stenosis location

rate
sampling rate in Hz

description
definition of automatic parameter sweep range

configuration
concrete parameters of the trapezoidal input curve (offset and amplitude in mmHg, ascend times and descend times and smoothing window in a fraction the time period (1.2s))

signalStruct

field
description

nodeId
corresponds to numbered nodes at which the sensor is placed, the corresponding location can be found in the technical paper describing the MACSim simulator (node numbering, not sensor numbers) or in the software SISCA in the example database.

type
'p' ... pressure or 'q' ... flow

data
double array, time series of each sensor,  unit mmHg for type 'p' and ml/s for type 'q'

anatomicalPosition
name of the corresponding anatomical position

Tools:

This Tools should make it easier to load the dataset. The usage is documented in the respective code files.

Code for the publication is available here:
https://gitlab.com/agbernhard.lse.thm/publication_macsim_machinelearning