Interpretable Supervised Machine Learning Models For Fetal Abnormalities Prediction
Emma E.Y Wilson, Micheal Francis Kalyango, Joyce Nakatumba Nabende · 2023
Prenatal treatment is complicated by fetal anomalies, thus timely intervention and better results depend on accurate prognosis. With the use of ultrasound, echocardiography, and magnetic resonance imaging (MRI), pregnant women and the developing fetus are medically monitored at certain stages. In this research, machine learning’s ability to understand and build methods that leverage data to improve computer performance on tasks is implemented on medical data of fetuses recorded during antenatal visits to identify certain features or factors that may lead to fetal abnormalities either during pregnancy or childbirth by using supervised learning techniques. The health data recordings were read using Cardiotocograms (CTGs), a simple and cost-accessible equipment. Ultrasound pulses were used, and the CTG was used to measure the reaction in order to determine the uterine contractions, fetal movements, FHR(fetal heart rate), and additional factors.Since AI models have demonstrated a high degree of precision in terms of their suitability for the health sector, comprehending the information these models rely on in order to make such correct decisions has become just as crucial as the models’ accuracy.This paper develops predictive models after analysing output of a Python library framework which is a machine learning(ML) approach that implements various algorithms with less user input in predicting fetal abnormalities. SHAP and LIME explainable AI (XAI) techniques were used for model accountability. Five trained classification models were developed and results assessed. LGBM classifier achieved a high accuracy of 96.71%, and GaussianNB of low accuracy 82.16% .