FHRMA Python

Replicating components of the MATLAB FHRMA package within Python

Context - features based on FIGO guidelines

In the context of the FIGO guidelines, the routinely judged FHR components, often called FIGO-based or morphological features, are the baseline heart rate, variability, accelerations, and decelerations. These components are the most robust indicators to ascertain fetal well-being. For this reason, in almost all studies that address automated CTG analysis, these basic morphological features are confirmed as an indispensable part of the analyses. The initial studies on automated CTG analysis focused on the detection of morphological features that clinicians examine with the naked eye, but this is challenging due to a lack of standards on how computers estimate the diagnostic indices.[source]

Studies have since explored the use of other features from linear, non-linear, discrete wavelet transform (DWT), empirical mode decomposition (EMD), time-frequency and image-based time-frequency (IBTF) domains.[source]

Refer to 03_literature_review in the ctg_exploratory repository for some of the publications describing this, as well as examples of commercial software for this, that have been evaluated in clinical practice.

Context - FHR baseline

Definition from the FIGO guidelines: "This is the mean level of the most horizontal and less oscillatory FHR segments. It is estimated in time periods of 10 minutes and expressed in beats per minute (bpm). The baseline value may vary between subsequent 10-minute sections."

Definition from the NICE guidelines: "Determine baseline fetal heart rate by looking at the mean fetal heart rate, excluding accelerations and decelerations, over a period of 10 minutes when the fetal heart rate is stable. When deciding if there is any change in baseline fetal heart rate, compare it with earlier CTG traces or recordings of fetal heart rate."

It is challenging to define a method for determining the FHR baseline because:

• It has a circular definition, as accelerations and decelerations are defined as periods when the signal is consistently above or below the baseline.
• There is high variability in some recordings, making it hard to distinguish between change in the baseline, and an acceleration or deceleration.[source]

M'Barek et al. 2023 state that you can overcome this issue by calibrating your method on baselines, accelerations and decelerations annotated by a consensus of obstetricians (they use those on the FHRMA dataset).[source] Wróbel et al. 2013 explain how Bernardes et al. 1996 "proposed to use additional information obtained from uterine contraction or fetal movement activity signals for removal particular signal segments from baseline calculation. However, significant subjectivity and high dependency on a signal recordings quality (using popular bedside monitors) are limiting this approach."

The FHRMA toolbox contains a range of published methods for FHR baseline calculation implemented within MATLAB.

FHRMA (Fetal Heart Rate Morphological Analysis)

Key sources:

• FHRMA github
• FHRMA wiki.
• Boudet, S., Houzé de l’Aulnoit, A., Demailly, R., Delgranche, A., Peyrodie, L., Beuscart, R., Houzé de l’Aulnoit,D. - Fetal heart rate signal dataset for training morphological analysis methods and evaluating them against an expert consensus. Preprints pp. Submitted to data in brief,2019, DOI:10.20944/preprints201907.0039.v1

License:

License: GPL-3.0. As from this summary, this license means:

1. Anyone can copy, modify and distribute this software.
2. You have to include the license and copyright notice with each and every distribution.
3. You can use this software privately.
4. You can use this software for commercial purposes.
5. If you dare build your business solely from this code, you risk open-sourcing the whole code base.
6. If you modify it, you have to indicate changes made to the code.
7. Any modifications of this code base MUST be distributed with the same license, GPLv3.
8. This software is provided without warranty.
9. The software author or license can not be held liable for any damages inflicted by the software.

Summary:

The dataset contains 155 FHR recordings in which a reference baseline, accelerations and decelerations have been annotated by expert consensus. 66 FHR recordings with a shared expert analysis have been included in a training dataset, and 90 other FHR recordings with a non-shared expert analysis have been included in an evaluation dataset. The dataset also contains the results produced by 12 re-coded automatic analysis methods from the literature.

Description:

The raw FHR records are available for the training and test data. These are .fhr files which can be opened with the fhropen.m function in MATLAB. They were sampled at 4Hz. For each sample, a uint32 serves as the UNIX timestamp for the beginning of the recording. Next, a uint16 corresponds to the FHR from the first sensor (multiplied by 4), a uint16 for the FHR signal from the second sensor (multiplied by 4) (unused here), a uint8 for the TOCO signal (multiplied by 2), and a uint8 reserved to store signal quality (unused here).

The main folder in the GitHub repository contains functions from the software, including functions for analysis with the 12 methods, which follow a file structure of 'aa'm then method - e.g. aamwmfb.m.

The analyses folder in the github contains MAT files which have record of when accelerations and decelerations occur for each record, according to the different methods. The methods are listed here. As described on the dataset page, these files contain:

• Filename (name of the record)
• Baseline (baseline signal sampled at 4Hz)
• Accelerations (2xn table with beginning and end of each acceleration in minutes)
• Decelerations (2xn table with beginning and end of each deceleration in minutes)
• Overshoots (2xn table with beginning and end of overshoots in minutes) (excluded from evaluation)
• Unreliable signal (periods of maternal heart rate or too much missing signal preventing good analysis) (excluded from evaluation)
• Not to analyse (period not to be analysed due to insufficient signal before or after for baseline positioning)
• Training data (boolean set to 1 if recording is part of the training dataset and 0 of part of evaluation)

The baseline, accelerations, decelerations, and over-shoots are not publicly shared for the recordings in the evaluation dataset to avoid any training on those data.

It states that the files are also available in WFDB format on the PhysioNet page, but I've not yet been able to identify the Physionet page.

These files are either from expert evaluation, or automated analysis from 12 different methods available in the literature. Methods in table below as from [source], ordered by their performance in their evaluation of them. Not included in FHRMA but included in the leaderboard are two methods from Zhong et al. 2022, which perform in 1st and 3rd place.

Function name	Filename	Publication	Baseline computation method	Acceleration/deceleration thresholding described?
aamwmfb	WMFB_orig.mat	Boudet et al. (2020)	Based on a weighted median filter baseline (WMFB). The weights are set to the prior probabilities that the FHR is in the baseline state. The unstable FHR periods have low probabilities. Successive baselines are computed with a progressive trimming process. Periods far from previous baselines have also low probabilities.	Yes
aamlu	L_std.mat	Lu et Wei (2012)	A method using empirical mode decomposition; periods exhibiting significant differences between two successive local FHR minima are excluded from the baseline computation.	No (probably standard thresholding)
aamhouze	H_std.mat	Houzé et al. (1990)	Based on a series of logic rules applied to various signal derivatives, followed by a smoothing step.	No (not described clearly enough to be reprogrammed)
aamtaylor	T_orig.mat / T_std.mat	Taylor et al. (2000)	A linear low-pass filtering method with progressive trimming.	Yes
aamwrobel	W_std.mat	Wróbel et al. (2013)	Based on myriad filters: an intermediate approach, between a linear average filter and use of the mode.	No (probably standard thresholding)
aammongelli	MG_std.mat	Mongelli et al. (1997)	This method computes a primary mode and a secondary mode; the two modes can be switched, depending on continuity and frequency criteria.	No
aamayres	A_std.mat	Ayres de Campos et al. (2000)	Based on “frequent” FHR values in a 10-min moving window.	Yes (= standard thresholding)
aammantel	MT_orig.mat / MT_std.mat	Mantel et al. (1990)	The method uses the mode or another local maximum of the FHR histogram to obtain an initial baseline. The final baseline is obtained by progressively trimming the first baseline (based on Dawes et al.’s method). This method was developed for antepartum recording.	Yes
aampardey	P_std.mat	Pardey et al. (2002)	The method first uses the histogram’s mode or another local maximum. The second baseline is obtained by linear filtering of the FHR and trimming as a function of the first baseline (based on Dawes et al.’s method).	Yes (=standard thresholding)
aamjimenez	J_orig.mat / J_std.mat	Jimenez et al. (2002)	Unstable periods (with a derivative above a cut-off) are removed, and the baseline is calculated after smoothing the remaining signal.	Yes
aammaeda	MD_std.mat	Maeda et al. (2012)	The baseline is set to the FHR level that occurs most often within a 10 bpm-wide bin.	No (not described clearly enough to be reprogrammed)
aamcazares	C_orig.mat/ C_std.mat	Cazares et al. (2002)	Based on morphological filters; an opening filter removes accelerations and a closing filter removes decelerations.	Only described for acceleration (presumably the same for deceleration)

It also contains work around identification of false signals which is described in this pre-print and in these python files.

In their presentation on YouTube, can see that the training and test data are from two different sources:

• Training dataset is 66 recordings, duration 90 min (30 min - 7 hours), digitial acquisition or scan, medium or hard to assess, consensus of 3 experts
• Test dataset is 90 recordings, duration 105 min (90 min - 120 min), digitial acquisition, easy or medium or hard to assess, 3 independent experts and a consensus

Citation:

The toolbox is related to several papers. Please cite those papers if you use any of the data or source code of this repository. [4] must be cited if you use the toolbox. [1] must be cited if you use the morphological analysis (baseline, Acceleration, deceleration) [3] must be cited if you use the morphological analysis dataset. [5] must be cited if you use the WMFB method (current best) for morphological analysis. [6] must be cited if you use the false signal detection, method, interface and/or dataset.

• [1] Houzé de l’Aulnoit, A., Boudet, S., Demailly, R., Delgranche, A., Peyrodie, L., Beuscart, R., Houzé de l’Aulnoit,D. - Automated fetal heart rate analysis for baseline determination and acceleration/deceleration detection: A comparison of 11 methods versus expert consensus. Biomedical Signal Processing and Control 49:113 -123,2019, DOI:10.1016/j.bspc.2018.10.002
• [2] Houzé de l'Aulnoit, Agathe, Boudet, Samuel, Demailly, Romain, Peyrodie, Laurent, Beuscart, Regis, Houzé de l'Aulnoit, Denis - Baseline fetal heart rate analysis: eleven automatic methods versus expert consensus. Engineering in Medicine and Biology Society (EMBC), 2016 IEEE 38th Annual International Conference of the pp. 3576--3581,2016, DOI:10.1109/EMBC.2016.7591501 Download on researchgate
• [3] Boudet, S., Houzé de l’Aulnoit, A., Demailly, R., Delgranche, A., Peyrodie, L., Beuscart, R., Houzé de l’Aulnoit,D. - Fetal heart rate signal dataset for training morphological analysis methods and evaluating them against an expert consensus. Preprints pp. Submitted to data in brief,2019, DOI:10.20944/preprints201907.0039.v1
• [4] Boudet, S., Houzé de l’Aulnoit, A., Demailly, R., Delgranche, A., Peyrodie, L., Beuscart, R., Houzé de l’Aulnoit,D. - A fetal heart rate morphological analysis toolbox for MATLAB. SoftwareX. 2020 Jan 1;11:100428. DOI:10.1016/j.softx.2020.100428
• [5] Boudet, S., Houzé de l’Aulnoit, A., Demailly, R., Peyrodie, L., Beuscart, R., Houzé de l’Aulnoit,D. - Fetal heart rate baseline computation with a weighted median filter. Computers in biology and medicine. 2019 Nov 1;114:103468. DOI:10.1016/j.compbiomed.2019.103468
• [6] Boudet, S., Houzé de l’Aulnoit, A., Demailly, R., Peyrodie, L., Houzé de l’Aulnoit,D. - Use of deep learning to detect the maternal heart rate and false signals on fetal heart rate recordings. Biosensors 2022; 12(9):691. DOI:10.3390/bios12090691