EP02.07: Novel algorithm of standard plane detection in second trimester fetal neurosonography using three‐dimensional automatic segmentation

Standard plane detection in three-dimensional (3D) fetal neurosonography is essential step in the assessment of fetal brain. The use of conventional methods for detecting standard plane from 3D volume are limited as success is highly affected by the direction of acquisition and the alignment of initial plane with seed points. To overcome these limitations, we developed a novel algorithm for detection of standard axial planes of fetal central nervous system (CNS) from 3D volume data sets using automatic segmentation and plane-navigation method. We used 3D volumes of the fetal head obtained from normal pregnancies at 20+0 to 28+6 weeks of gestation. The volumes were acquired using a HERA W10 ultrasound system (Samsung Medison, Seoul, Republic of Korea) equipped with a 1-8 MHz transabdominal probe. Our proposed method automatically segmented the 3D volume data sets and categorised them into: thalamus, cavum septum pellucidum, cerebellum, choroid plexus, posterior ventricle cavity, cisterna magna and midline falx using 3D segmentation algorithm. The categorised segments were then used to navigate and identify the three standard axial planes (transventricular (TV), transthalamic (TT) and transcerebellar (TC) plane). The output standard plane images were audited by two independent experts to confirm plane accuracy. In total, 500 3D volumes were obtained from 537 patients. The experimental results showed that our method achieved a success rate of 93% for 3D autosegmentation and accuracy of 93%, 94%, and 94% for TV, TT, and TC plane detection, respectively. In addition, our method successfully detected standard axial planes within the acquired volume irrespective of the direction of volume acquisition and tilt or rotation of the transducer. Our algorithm showed more effective and higher performance than conventional software in detecting the standard axial planes of fetal CNS.