MR imaging of the placenta affords improved resolution and large fields of view, but is currently limited by subjective, qualitative clinical impressions. Our objective was to develop artificial neural networks (a deep learning AI method) to automate MR volume and 3-dimensional display of placental segmentation as the first step in quantification of placental pathology. We selected 100 MR cases performed between 20 and 38 weeks’ gestation from 2012-19 for suspected fetal abnormalities but in which no major abnormality was identified. An expert, serving as ground truth, and learner segmented the uterus and placenta from axial 7-mm single shot fast spin echo sequences. A customized version of U-Net was developed for automated segmentation, using a high performance computer workstation, with 5 points manually selected to identify boundaries of the uterus and placenta, as well as the approximate center of the placenta. The neural network trained with 70 MR cases and validated on 10, followed by independent testing of 20 cases. The performance of the algorithm was evaluated by Dice similarity coefficient (DSC). The learner achieved a DSC of 0.92±0.07 for placenta (N=40) and 0.95±0.02 for uterus (N=25), compared to deep learning-based algorithm DSC of 0.82±0.06 for placenta and 0.92±0.04 for uterus (Table 1). The processing time was 22 seconds, compared to an average of 30 minutes for the learner. Figure 1 is a 3-D rendered automated segmentation of the placenta and uterus.Tabled 1Table 1UterusPlacentaDeep LearningNDSC (%)V (cm3)ΔV (%)DSC (%)V (cm3)ΔV (%)Training7095.9 ± 1.92485 ± 8500 ± 890.1 ± 4.6737 ± 2687 ± 9Validation1092.3 ± 5.12071 ± 6522 ± 880.9 ± 8.3539 ± 2116 ± 28Test2092.1 ± 4.02411 ± 8151 ± 982.0 ± 5.9589 ± 263-6 ± 17 Open table in a new tab We developed a deep learning AI method of automatic segmentation of the uterus and placenta with MR rendered in 3-D. This first step allows for future MR parametric analysis within that segmentation and quantitative assessment of placental pathology, such as placenta accreta spectrum.View Large Image Figure ViewerDownload Hi-res image Download (PPT)