2018/04/24 by Matthew Sinclair, Sinclair, Matthew, Christian F. Baumgartner +21 · 3 citations
Biochemistry, Genetics and Molecular Biology · Computer Science · Medicine · #Cleft Lip and Palate Research #Computer Vision and Pattern Recognition (cs.CV) #Domain Adaptation and Few-Shot Learning #FOS: Computer and information sciences #Fetal and Pediatric Neurological Disorders
paper · pdf · doi:10.48550/arxiv.1804.09102
openalex publication_date 2018/04/24 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Measurement of head biometrics from fetal ultrasonography images is of key\nimportance in monitoring the healthy development of fetuses. However, the\naccurate measurement of relevant anatomical structures is subject to large\ninter-observer variability in the clinic. To address this issue, an automated\nmethod utilizing Fully Convolutional Networks (FCN) is proposed to determine\nmeasurements of fetal head circumference (HC) and biparietal diameter (BPD). An\nFCN was trained on approximately 2000 2D ultrasound images of the head with\nannotations provided by 45 different sonographers during routine screening\nexaminations to perform semantic segmentation of the head. An ellipse is fitted\nto the resulting segmentation contours to mimic the annotation typically\nproduced by a sonographer. The model's performance was compared with\ninter-observer variability, where two experts manually annotated 100 test\nimages. Mean absolute model-expert error was slightly better than\ninter-observer error for HC (1.99mm vs 2.16mm), and comparable for BPD (0.61mm\nvs 0.59mm), as well as Dice coefficient (0.980 vs 0.980). Our results\ndemonstrate that the model performs at a level similar to a human expert, and\nlearns to produce accurate predictions from a large dataset annotated by many\nsonographers. Additionally, measurements are generated in near real-time at\n15fps on a GPU, which could speed up clinical workflow for both skilled and\ntrainee sonographers.\n