Deep Learning-Based Fetal Health Classification: A Comparative Analysis of Convolutional and Recurrent Neural Networks

Abstract

Fetal health monitoring plays a crucial role in prenatal care, enabling early detection of complications that may impact pregnancy outcomes. Traditional methods, including cardiotocography (CTG), rely on expert interpretation, which can introduce variability and potential misdiagnoses. In this study, deep learning techniques are employed to classify fetal health conditions based on CTG data. A comparative analysis is conducted on six architectures: Multilayer Perceptron (MLP), Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Bidirectional LSTM (BiLSTM), and Attention-based LSTM.  The models are evaluated using accuracy, precision, recall, and F1-score under a 10-fold cross-validation framework. Results indicate that CNN outperforms all other models, achieving an accuracy of 97.18% due to its hierarchical feature extraction capabilities. GRU demonstrates competitive performance with an F1-score of 95.50% while maintaining computational efficiency. The study further includes a complexity analysis, revealing that recurrent models, particularly BiLSTM and Attention-LSTM, introduce significant computational overhead without yielding substantial performance gains.  Potential threats to validity, including dataset bias and overfitting, are analyzed to ensure robust findings. The insights gained from this research highlight the advantages of CNN-based architectures in automated fetal health assessment and suggest future work integrating hybrid models and explainable AI techniques. These findings contribute to advancing AI-driven fetal monitoring systems, aiding clinical decision-making, and improving perinatal care.