MACHINE LEARNING BASED CLINICAL DECISION SUPPORT SYSTEM TO PREDICT FETAL HYPOXIA IN WOMEN DURING ANTENATAL CHECK-UP.

Abstract

BACKGROUND:Most under-five deaths occur within the first month after birth and intrapartum complications are a major contributor to the cause of death. These defects can be easily identified during the ante-natal check-up by use of a non-stress test. Due to the lack of availability of resources and medical experts in remote areas clinical decision support systems powered by machine learning models can provide information to the healthcare provider to make timely and better-informed decisions based on which course of treatment can be planned. AIM:The study aims to develop an accurate and sensitive clinical decision support system model that can identify pathological fetuses based on the fetal heart rate recordings taken during the non-stress test. METHOD: Foetal Heart rate recordings along with 10 other variables were collected from 1800 pregnant women in their third trimester. The data was put through a feature selection algorithm to identify important variables in the set. The data set was randomly divided into 2 independent random samples in the ratio of 70% for training and 30% for testing. After testing various machine learning algorithms based on specificity, sensitivity to accurately classify the fetus into normal, suspected, or pathological Random Forest algorithm was chosen. RESULT:The fetal status determined by Obstetrician 77.85% observations from the normal category, 19.88% from the suspected category, and 8.28% from the pathological category. The Boruta algorithm revealed that all 11 independent variables in the data set were important to predict the outcome in the test set. In the training set the model had an accuracy of 99.04% and in the testing set accuracy was 94.7% (p-value=< 2.2e-16) with the precision of 97.56% to detect the pathological category. CONCLUSION:With the ability of the model to accurately predict the pathological category the CDS can be used by healthcare providers in remote areas to identify high-risk pregnant women and take the decision on the medical care to be provided.