PresentationPDF Available

Abstract

Automatic screening of the population for congestive heart failure (CHF) is a matter of pressing concern due to the severity of the health consequences resulting in disability and death of people. On the one hand, portable devices working with ECG signals become convenient tools for the lay user due to the simplicity. On the other hand, analyzing the specific behavior of the R-peaks sequence (analysis of heart rate variability) in cardiac pathologies allows identifying the patterns inherent in particular heart dysfunction. Such patterns are effectively differentiated using symbolic dynamics methods and the subsequent application of machine learning methods. In this study, a highly specific model was obtained (sensitivity 0.71, specificity 0.96), suitable for automatic screening of CHF. Its operability and performance characteristics have been verified through testing in several publicly available databases.
A.A. Pulavskyi, S.S. Krivenko, S.A. Krivenko, I.V. Linskiy, M.F. Posokhov, L.S. Kryvenko
Scientific Company KOLIBRI LLC, Kharkiv National University of radioelectronics, Institute of Neurology, Psychiatry
and Narcology of the National Academy of Medical Sciences of Ukraine, Kharkiv National Medical University
Automatic recognition of
congestive heart failure signs
in heart rate variability data
presented at MECO’2022 and CPSIoT’2022, Budva, Montenegro
www.mecoconference.me
Outline
PROBLEM DEFINITION
HANDCRAFTED FEATURES
DATABASES AND PREPROCESSING
MODEL AND DISCUSSION OF RESULTS
CONCLUSIONS
MECO&CPSIoT’2022, Budva, Montenegro
Problem definition
Cardiovascular diseases are still the leading cause of death and
disability according to WHO
Using of telemedicine devices by laypersons poses new
challenges for their developers
When working with an ECG signal, the analysis of the R-peak
sequence comes to the fore (IBI - interbeat interval sequence)
The division of patterns into physiologically normal and
pathological becomes more accurate since it is based on heart
rate variability (HRV) originated from IBI analysis
The purpose of this study was to create a highly specific
diagnostic model capable of automatically identifying patients
with congestive heart failure (without the involvement of a
professional)
MECO&CPSIoT’2022, Budva, Montenegro
Handcrafted features
Heart rate variability is calculated as the time difference of
successive heartbeats on an ECG. The optimum is the
difference between the P-peaks, but since the P-peak is often
poorly differentiable and subject to noise, the R-peak
difference is considered. That is the time component of HRV.
To find differentiating characteristics, we have used five
symbolic analysis methods based on HRV:
Ordinal patterns (op)
Delta coding (dc)
Binary delta coding (dcbin)
Sigma coding (sc)
Max-min method (mmc)
MECO&CPSIoT’2022, Budva, Montenegro
Signal Preprocessing
The preprocessing of the ECG segment included:
Notch filters at 50 and 60 Hz
Bandpass filter of 0.545 Hz
A standard QRS detector has been used to obtain R-
peaks (raw IBI)
MECO&CPSIoT’2022, Budva, Montenegro
Databases
Kaggle ECG Database
BIDMC Congestive Heart Failure Database
Congestive Heart Failure RR Interval Database (CHFRR)
MIT-BIH Normal Sinus Rhythm Database (MBN)
Combined measurement of ECG, Breathing and Seismocardiograms
(CEBS)
Fantasia Database
PTB Diagnostic ECG Database
MECO&CPSIoT’2022, Budva, Montenegro
Results
MECO&CPSIoT’2022, Budva, Montenegro
Databases
Test set MBN+CHFRR CEBS Fantasia PTB
Metric
Sen Spec Sen Spec Spec Spec Spec
Model
0.71 0.99 0.86 0.96 1.00 0.97 0.78
Conclusions
A highly specific LightGBM model of automatic
screening for congestive heart failure was obtained
during the study.
The model's sensitivity was not less than 0.71, and the
specificity was not less than 0.95 on test databases.
Automatic screening implies the determination of
signs of the disease without manual error correction
of RR intervals, including the correction of ectopic
beats, long or short beats, missed beats, extra beats.
The proposed approach can be used for personal
telemedicine devices in the early detection of health
threats.
MECO&CPSIoT’2022, Budva, Montenegro
THANK YOU
Q&A
Liudmyla Kryvenko
info@kolibri.one
MECO&CPSIoT’2022, Budva, Montenegro
ResearchGate has not been able to resolve any citations for this publication.
ResearchGate has not been able to resolve any references for this publication.