Automated Maternal–Fetal Health Analysis through Deep Neural Network Integration in Ultrasound Imaging

Ultrasonography is a non-invasive and widely adopted imaging modality for monitoring fetal development and detecting potential complications during pregnancy. It provides essential information on fetal biometric measurements, anatomical structures, fetal movements, and growth abnormalities. This study addresses two tasks: (i) fetal biometric parameter-based classification (four categories) and (ii) trimester-based classification (three categories). Two datasets are employed: the publicly available HC-18 dataset and a custom dataset containing Head Circumference (HC), Femur Length (FL), Abdominal Circumference (AC), and Crown-Rump Length (CRL). Biometric parameters served as features for parameter-based classification, while HC, FL, and HC-18 are used for trimester-based classification into First Trimester (FT), Second Trimester (ST), and Third Trimester (TT). Feature extraction is performed using pre-trained deep learning models (ResNet50, VGG16, VGG19, and Xception), and a feature fusion strategy is applied to combine their representations. Gradient-weighted Class Activation Mapping (Grad-CAM) is incorporated to identify regions of interest (ROI) and improve model interpretability. The framework is evaluated using standard performance metrics, including accuracy, precision, recall, and F1-score. The parameter-based classification achieved an accuracy of 99.65%, while the trimester-based classification achieved 96.24% for HC, 97.70% for FL, and 93.33% for HC-18. The experimental findings demonstrate that the proposed feature-fusion framework achieves robust and reliable classification performance for fetal ultrasound analysis. The integration of Grad-CAM further enhances interpretability, making the method a comprehensive solution for the automated analysis of two-dimensional fetal ultrasound images.