EP01.17: Assessing the feasibility of three‐dimensional auto segmentation‐based standard plane detection for fetal brain malformations: a preliminary study

Standard plane detection in three-dimensional (3D) fetal neurosonography was essential in assessing the fetal brain. Our proposed algorithm, standard axial plane detection based on 3D auto segmentation and anatomical knowledge-based plane navigation, demonstrated higher performance than conventional software in detecting the standard axial planes of the fetal normal central nervous system. However, research on algorithms accurately detecting standard planes in fetal brain malformations was limited. This study aimed to evaluate the performance of the proposed standard plane detection algorithm in fetal brain malformations. We used 29 3D volumes of the fetal head obtained from fetal brain malformation. The volumes were acquired using a HERA W10 ultrasound system (Samsung Medison, Seoul, Republic of Korea) with a 1-8 MHz transabdominal probe. Two independent experts audited the output standard axial plane (transventricular (TV), transthalamic (TT), and transcerebellar (TC) plane) images to confirm diagnostic plane accuracy. The experimental results showed that our method achieved a success rate of 79.3% for 3D auto-segmentation and accuracy of 75.9%, 65.5%, and 72.4% for TV, TT, and TC plane detection, respectively. In cases of mild fetal brain anomalies such as ventriculomegaly and cerebellar hypoplasia, the accuracy rate reached 81.3%. In cases of corpus callosal dysgenesis, the detection rates for the TV, TT, and TC planes were 75%, 62.5%, and 75%, respectively. However, severe hydrocephalus, characterised by unclear internal structures, rendered plane detection impossible. Our proposed algorithm demonstrated good performance in fetal brain anomalies, excluding severe hydrocephalus. Techniques that can acquire standard planes even in malformations where the brain's internal structure is complicated to distinguish are needed.