Deep Learning Based Fetal Face Detection And Visualization In Prenatal Ultrasound

Accurate detection and visualization of fetal face is crucial in prenatal diagnosis and detection of any fetal anomalies. In obstetrics scans, 3D ultrasound is now routinely used, however detection of optimal plane for viewing the fetal face is cumbersome and time consuming for the trained specialist and sonographer. In this paper, we propose a novel fetal face detection and visualization approach using 3D ultrasound volumes. The novelty is in the approach for training a deep learning network for fetal face detection, segmentation and visualization. Given a 3D ultrasound volume, the proposed solution automatically segments fetal facial surface, key landmarks and detects the position and orientation of these structures. The findings indicate that our approach has excellent detection accuracy not only in case of single fetus but also in multiple fetus pregnancy such as twins or triplets.