Fetal Abdominal Structures Segmentation Dataset Using Ultrasonic Images

Between September 2021 and September 2023, an extensive dataset of fetal ultrasound images was meticulously compiled. The study featured 169 subjects, each contributing a variable number of fetal abdomen circumference (AC) images, totaling nearly 1500 images.

Eligible participants were term pregnant women aged 18 years or older, either in labor or scheduled for delivery at the Maternity of University Hospital Polydoro Ernani de São Thiago in Florianópolis, Santa Catarina, Brazil. This encompassed patients scheduled for labor induction, cesarean section, and those facing pregnancy complications, such as the rupture of membranes, gestational diabetes, pre-eclampsia, and intrauterine growth restriction.

Non-eligible cases included pre-term pregnancies, multiple pregnancies, and pregnancies with fetal structural or chromosomal anomalies. The gestational age for all pregnancies was determined based on the earliest ultrasound examination.

The study received ethical approval from the Ethics Committee in Human Research at the authors' institution, UNIVERSIDADE FEDERAL DE SANTA CATARINA – UFSC, with approval number 4.971.754. All eligible participants were informed about the study's objectives and provided written consent.

Expert clinicians used a standardized acquisition protocol to obtain ultrasound images. A common axial section of the fetal AC was captured, with measurements taken at the widest part of the fetal abdomen, spanning the liver. This section encompassed the fetal stomach, aorta artery, spine, and intrahepatic portion of the umbilical vein.

Ultrasound images were acquired using Siemens Acuson, Voluson 730 (GE Healthcare Ultrasound), or Philips-EPIQ Elite (Philips Healthcare Ultrasound) equipment, equipped with 2–9 MHz curved linear transducers. For validation, the ultrasound presets excluded post-processing options such as image smoothing, zoom, calipers, pointers, or Doppler measurements. Tissue harmonic imaging and adjustments to settings like frequency, gain, and focus were at the discretion of the performing physician. Images were digitally stored in the original Digital Imaging and Communication in Medicine (DICOM) format for offline analysis.

A rigorous quality control process, overseen by two team members, ensured that images with calipers or evident acoustic shadows from bony structures were excluded. Only images meeting these strict quality criteria proceeded to the next phase.

Expert clinicians utilized specialized software, specifically version 5.2.2 of 3D Slicer, to process selected DICOM images. This software streamlined the segmentation of abdominal structures, emphasizing key components like the abdominal aorta artery, intrahepatic umbilical vein, stomach, and maximizing liver area for in-depth analysis.