Automatic fetal face detection by locating fetal facial features from 3D ultrasound images for navigating fetoscopic tracheal occlusion surgeries

With the wide clinical application of 3D ultrasound (US) imaging, automatic location of fetal facial features from US volumes for navigating fetoscopic tracheal occlusion (FETO) surgeries becomes possible, which plays an important role in reducing surgical risk. In this paper, we propose a feature-based method to automatically detect 3D fetal face and accurately locate key facial features without any priori knowledge or training data. The candidates of the key facial features, such as the nose, eyes, nose upper bridge and upper lip are detected by analyzing the mean and Gaussian curvatures of the facial surface. Each feature is gradually identified from the candidates by a boosting traversal scheme based on the spatial relations between each feature. In experiments, all key feature points are detected for each case, and thus a detection success rate of 100% is achieved by using 72 3D US images from a test database of 6 fetal faces in the frontal view and any pose within 15° from the frontal view, and the location error 3.18 ± 0.91 mm of the detected upper lip for all test data is obtained, which can be tolerated by the FETO surgery. Moreover, this system has a high efficiency and can detect all key facial features in about 625 ms on a quad-core 2.60 GHz computer.