Title

Classification of Foetal Distress and Hypoxia Using Machine Learning Approaches

Source of Publication

Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

Abstract

© 2018, Springer International Publishing AG, part of Springer Nature. Foetal distress and hypoxia (oxygen deprivation) is considered as a serious condition and one of the main factors for caesarean section in the obstetrics and Gynecology department. It is the third most common cause of death in new-born babies. Many foetuses that experienced some sort of hypoxic effects can develop series risks including damage to the cells of the central nervous system that may lead to life-long disability (cerebral palsy) or even death. Continuous labour monitoring is essential to observe the foetal well being. Foetal surveillance by monitoring the foetal heart rate with a cardiotocography is widely used. Despite the indication of normal results, these results are not reassuring, and a small proportion of these foetuses are actually hypoxic. In this paper, machine-learning algorithms are utilized to classify foetuses which are experiencing oxygen deprivation using PH value (a measure of hydrogen ion concentration of blood used to specify the acidity or alkalinity) and Base Deficit of extra cellular fluid level (a measure of the total concentration of blood buffer base that indicates the metabolic acidosis or compensated respiratory alkalosis) as indicators of respiratory and metabolic acidosis, respectively, using open source partum clinical data obtained from Physionet. Six well know machine learning classifier models are utilised in our experiments for the evaluation; each model was presented with a set of selected features derived from the clinical data. Classifier’s evaluation is performed using the receiver operating characteristic curve analysis, area under the curve plots, as well as the confusion matrix. Our simulation results indicate that machine-learning algorithms provide viable methods that could delivery improvements over conventional analysis.

Document Type

Conference Proceeding

ISBN

['9783319959566']

First Page

767

Last Page

776

Publication Date

1-1-2018

DOI

10.1007/978-3-319-95957-3_81

Share

COinS