Automating first-trimester placenta volume measurements in three-dimensional ultrasonography using virtual reality and artificial intelligence: The Rotterdam Periconception Cohort.
Authors
Affiliations (4)
Affiliations (4)
- Department of Obstetrics and Gynecology, Erasmus MC, Rotterdam, Netherlands; Biomedical Imaging Group Rotterdam,Department of Radiology and Nuclear Medicine, Erasmus MC, Rotterdam, Netherlands. Electronic address: [email protected].
- Department of Obstetrics and Gynecology, Erasmus MC, Rotterdam, Netherlands.
- Department of Pathology, Erasmus MC, Rotterdam, Netherlands.
- Biomedical Imaging Group Rotterdam,Department of Radiology and Nuclear Medicine, Erasmus MC, Rotterdam, Netherlands.
Abstract
Volumetric measurements of the placenta in the first trimester provide insights into placental development and early markers for placenta-related complications. Current manual placental volume (PV) measurement using Virtual Organ Computer-Aided Analysis (VOCAL) on three-dimensional (3D) ultrasound is time-consuming and limited to two-dimensional (2D) planes. We aimed to validate a Virtual Reality (VR)-based 3D PV measurement and develop an automated Artificial Intelligence (AI) method for PV measurement. 3D ultrasound images from 83 singleton pregnancies (115 images) at 7, 9, and 11 weeks gestation from the Rotterdam Periconception Cohort were analyzed. A validation set of 414 paired images was added to assess AI-PV reproducibility using two sequential 3D ultrasounds. VOCAL-PV involved manual tracing at 15° intervals in 2D planes. VR-PV used an in-house VR system for manual voxel selection in 3D. AI-PV used nnU-Net trained on VR-PV. Absolute error, Dice scores, and measurement time were compared. VR-PV showed average absolute errors of 2.0 cm<sup>3</sup> (week 7), 4.4 cm<sup>3</sup> (week 9), and 7.6 cm<sup>3</sup> (week 11). AI-PV had similar errors respectively: 2.2 cm<sup>3</sup>, 4.5 cm<sup>3</sup>, and 9.2 cm<sup>3</sup>. Dice overlap between AI-PV and VR-PV was 79-82%. AI-PV reproducibility errors were respectively 0.7 cm<sup>3</sup>, 1.2 cm<sup>3</sup>, and 1.2 cm<sup>3</sup>. AI-PV measurement time was 1 min versus 11 min for VOCAL and VR. VR-PV and AI-PV enable accurate first-trimester PV measurement. AI-PV eliminates manual input, reduces time, and supports scalable PV assessment for research and early clinical screening, with potential to improve early detection and management of placenta-related complications.