Early Detection of Fetal Abnormalities Using BiLSTM: A Deep Learning Framework for Prenatal Care

Detecting fetal abnormalities at an early stage enables limited risks to newborns and enhances prenatal medical care. CTG data contains vital information about fetal welfare which requires manual interpretation by clinicians whose assessments show inconsistent results between practitioners. The proposed method applies deep learning through BiLSTM to automatically discover fetal abnormalities in CTG recordings by analyzing their surface information. To ensure data quality, Z-score normalization is applied for feature standardization, preventing scale variations from affecting model performance. The Synthetic Minority OverSampling Technique (SMOTE) addresses class imbalance by generating synthetic samples, improving model generalization for minority classes. Additionally, Recursive Feature Elimination (RFE) is employed to identify the most significant features, reducing model complexity and enhancing efficiency. A BiLSTM model works on the Fetal Health Dataset to process clinical data for classifying fetal states by normal, suspected and pathological conditions. The evaluation measures include Accuracy, Precision, Recall, F1-score, together with Area Under the Curve (AUC-ROC). Our BiLSTM model reaches experimental results of $\mathbf{9 3. 6} \%$ accuracy with $\mathbf{9 1. 8} \%$ precision and $92.2 \%$ recall along with $92.0 \%$ F1-score and $93.12 \%$ Specificity and an AUC-ROC of $95.1 \%$ above traditional machine learning classifiers. These findings highlight the potential of deep learning-driven diagnostic tools in prenatal healthcare. The proposed BiLSTM framework provides an automated, efficient, and interpretable approach to fetal health classification, aiding clinicians in timely and accurate decisionmaking. Future work will explore explainability methods and real-time deployment to further enhance clinical applicability.