A Deep CNN-Based Pipeline for Automated Classification of Anatomical Fetal Brain Views in Prenatal Ultrasound Imaging

Abstract

Fetal brain classification in prenatal ultrasound imaging is essential for detecting congenital anomalies and developmental disorders at an early stage. Accurate identification of anatomical planes supports clinicians in making timely diagnostic decisions. In this study, a labeled dataset of ultrasound images was used, covering four standard categories: Trans-cerebellum, Trans-thalamic, Trans-ventricular, and Other. A deep learning pipeline was developed using a pre-trained VGG19 CNN model with transfer learning for feature extraction. The architecture was fine-tuned by applying global average pooling and a softmax classifier. The system was trained with categorical cross-entropy loss and optimized using Adam. Results confirmed high classification accuracy, indicating the potential of automated tools to standardize imaging and reduce diagnostic variability in prenatal care.