Fully Automated 3-D Ultrasound Segmentation of the Placenta, Amniotic Fluid, and Fetus for Early Pregnancy Assessment

Abstract

Volumetric placental measurement using 3D ultra-1 sound has proven clinical utility in predicting adverse pregnancy 2 outcomes. However, this metric can not currently be employed 3 as part of a screening test due to a lack of robust and real-time 4 segmentation tools. We present a multi-class convolutional neural 5 network (CNN) developed to segment the placenta, amniotic fluid 6 and fetus. The ground truth dataset consisted of 2,093 labelled 7 placental volumes augmented by 300 volumes with placenta, 8 amniotic fluid and fetus annotated. A two pathway, hybrid 9 model using transfer learning, a modified loss function and 10 exponential average weighting was developed and demonstrated 11 the best performance for placental segmentation, achieving a 12 Dice similarity coefficient (DSC) of 0.84 and 0.38 mm average 13 Hausdorff distance (HDAV). Use of a dual-pathway architecture, 14 improved placental segmentation by 0.03 DSC and reduced 15 HDAV by 0.27mm when compared with a naïve multi-class model. 16 Incorporation of exponential weighting produced a further small 17 improvement in DSC by 0.01 and a reduction of HDAV by 18 0.44mm. Per volume inference using the FCNN took 7-8 seconds. 19 This method should enable clinically relevant morphometric 20 measurements (such as volume and total surface area) to be 21 automatically generated for the placenta, amniotic fluid and 22 fetus. Ready availability of such metrics makes a population-23 based screening test for adverse pregnancy outcomes possible.

Toward deep observation: A systematic survey on artificial intelligence techniques to monitor fetus via ultrasound images

“…An image segmentation task was used in (n= 4, 3.73%) studies for the purpose of fetal mage segmentation was target task anatomical structures identification. In ( Looney et al., 2021 ; Yang et al., 2019 ), 3D US were utilized to segment fetus, gestational sac amniotic fluid, and placenta in the beginning of the second trimester. However, in ( Li et al., 2017 ), 2D US was used to segment amniotic fluid and the fetus in the late trimester.…”

Amniotic Fluid Classification and Artificial Intelligence: Challenges and Opportunities

“…The results of the proposed model were best achieved using the RF classifier, which obtained a DSC of 0.876 and pixel accuracy of 0.857. Furthermore, Looney et al [19] proposed a multiclass CNN model to segment the placenta, AF, and fetus. The dataset consisted of 2093 labeled placental volumes augmented by 300 volumes with placenta, AF, and fetus annotated for multi-class segmentation.…”

Placenta segmentation in ultrasound imaging: Addressing sources of uncertainty and limited field-of-view

“…Convolutional neural networks (CNNs) have recently become the state-of-the-art tools for accurate segmentation ( Wang et al, 2020 ). When a large amount of labeled training data is available, supervised CNN approaches show impressive performance in a variety of medical image segmentation tasks, including good performances for segmenting the placenta in 3D US images ( Looney et al, 2018 ; Yang et al, 2019 ; Torrents-Barrena et al, 2019a ; Zimmer et al, 2019 , 2020 ; Schwartz et al, 2022 ; Looney et al, 2021 ). One major drawback is, however, that accurate expert pixel-level annotations are expensive and time-consuming to acquire.…”