Proceedings of the IEEE 15th International Symposium on Biomedical Imaging (ISBI 2018) - Deep learning with ultrasound physics for fetal skull segmentation

Congreso

Resumen *

Two-dimensional ultrasound (2DUS) is the primary screening modality for fetal screening, thanks to the lack of harmful effects on the fetus and mother, relatively low cost, and possibility of real time imaging and reporting. Nowadays, a second trimester US examination is routinely performed at 18?20 weeks of gestation in most countries, including a fetal survey, and a complete anatomical screening examination. As part of this comprehensive analysis, 2DUS-based biometry (i.e., sonographic measurements of the fetal anatomy) has been extensively used to indirectly assess the growth and well-being of the fetus, and estimating fetal weight [1]. The assessment of the fetal skull is an essential part of routine sonographic examination, including head circumference (HC), biparietal diameter (BPD), and occipitofrontal diameter (OFD). However, these biometrics are prone to errors, still relying on 2D measurements manually extracted from a specific anatomical planes (i.e., the transthalamic or transventricular planes). This subjectivity and operator dependence can affect the diagnostic capability of US-based fetal screening, limiting reproducibility, and may hinder the early detection of malformations [2]. Moreover, the early detection of cranial malformations, such as dolichocephaly, or brachycephaly, requires the detail structural analysis of the skull, its curvilinear bones, and boundaries, which may be difficult to visualize and quantify in a single 2D plane. Información suministrada por el agente en SIGEVA