Ultrasound images datasets
This research was a retrospective multicenter diagnostic study. For AI
model development and testing, abnormal pregnancies of 12 types of
common CNS malformations and normal pregnancies were retrospectively
collected from The first Affiliated Hospital of Sun Yat-sen University
(March 2010 to September 2018), Dongguan Maternal and Child Health
Hospital (January 2016 to December 2018), and the Women and Children’s
Hospital affiliated with Xiamen University (January 2016 to December
2018). These 12 types of malformations included: agenesis of corpus
callosum (ACC), absence of cavum septi pellucidi (ASP),
holoprosencephaly (HPE), Dandy-Walker malformation and variant (DWNv),
Megacisterna magna (MCM), Blake’s pouch cyst, hydrocephaly,
ventriculomegaly, arachnoid cyst, choroid plexus cyst (CPC), midline
cyst and subependymal cyst. All the prenatal ultrasonic diagnoses were
confirmed by prenatal or postnatal MRI, follow-up examination or
autopsy. Ultrasound examinations of the abnormal pregnancies over a
period of four weeks were included as part of this study. The mean
gestational age was 21+5 weeks and 25+4 weeks for normal and abnormal
cases, respectively. Ultrasound examinations were performed using
various machines from six different manufacturers (GE Voluson 730
Expert/E6/E8/E10, Aloka SSD-a10, Siemens Acuson S2000, Toshiba XARIO 200
TUS-X200, Samsung UGEO WS80A, Philips IU22). This retrospective study
was approved by Institutional Review
Board of The First Affiliated Hospital of Sun Yat-sen University.
Informed consent from patients was waived because of the retrospective
nature of the study.
Two-dimensional neurosonographic grayscale images were employed to
develop and testing the AI system. If the images were 3D volume data or
were with split-view, we would export it or divide it into qualified
single two-dimensional grayscale images before use according to the
methods introduced in our previously published study21. All the two-dimensional grayscale images should
meet the following criteria of inclusion: 1) neurosonographic images of
the standard axial planes, namely the transventricular (TV) plane,
transthalamic (TT) plane or transcerebellar (TC) plane, acquired
according to the guidelines of the International Society of Ultrasound
in Obstetrics & Gynecology (ISUOG) 22,23; 2) images with an integrated skull, properly
magnified without measurement caliper overlays and without the obvious
acoustical shadow. Consequently, after excluding unqualified images and
redundant normal images in the test dataset at Xiamen hospital, the
overall dataset contained 20,689 normal images and 17,573 abnormal
images. The pixel sizes of images were 1920 × 1080, 1408 × 712, 1400 ×
700, 1300 × 870, 960 × 720, 800 × 600, 768 × 576, 720 × 576 and 640 ×
480. The detailed constitutions of the ultrasound image datasets for the
development and testing of the AI system are shown in Table 1, and the
workflow diagram is shown in Figure 1.