Towards the early detection of Fetal Health using deep learning

Abstract

Fetal health detection plays a very important role in prenatal care, as it aids in early identification of potential health issues for both the mother and baby. Traditional diagnostic methods, such as cardiotocography (CTG) and ultrasound, face limitations in terms of accuracy and sensitivity to subtle anomalies. This research aims to develop a deep learning-based system using Feedforward Backpropagation Neural Networks (FBNNs) for classifying fetal health conditions into three categories: Normal, Suspect, and Pathological. The system utilizes clinical data from the “Fetal Health Classification” dataset obtained from Kaggle, which includes key features like fetal heart rate patterns and uterine contractions. Various activation functions, including ReLU, PReLU, and sigmoid, were tested, along with optimizers such as Adam, RMSProp, and SGD. The best results were achieved using Model 1, which combined ReLU in the hidden layers and Sigmoid in the output layer, resulting in high accuracy and performance. The model demonstrated its potential to overcome the limitations of traditional methods by offering a scalable, reliable, and efficient tool for early detection of fetal health conditions. This study contributes to advancing the use of AI in healthcare, particularly in improving prenatal care and enabling timely medical interventions.