Evaluation of Spatial Attentive Deep Learning for Automatic Placental Segmentation on Longitudinal MRI

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Obstetrics

Background:

• Automated placental segmentation using MRI in early pregnancy aids in predicting placental function and pregnancy outcomes.
• Segmentation methods effective at one gestational age may not generalize to others.

Purpose of the Study:

• To evaluate a spatial attentive deep learning method (SADL) for automated placental segmentation on longitudinal placental MRI scans.
• To assess the performance of SADL across different gestational ages.

Main Methods:

• A prospective, single-center study included 154 pregnant women scanned at 14-18 and 19-24 weeks of gestation.
• T2-weighted T2-HASTE MRI sequences were used.
• Manual segmentation by a clinical fellow served as the reference standard.
• Spatial attentive deep learning (SADL) and U-Net were compared using the Dice similarity coefficient (DSC).

Main Results:

• SADL achieved significantly higher average DSCs (0.83 ± 0.06 and 0.84 ± 0.05) compared to U-Net (0.77 ± 0.08 and 0.76 ± 0.10) at both gestational age ranges.
• Volume measurement differences between SADL and manual segmentation were within 95% limits of agreement for 90.4% of scans.

Conclusions:

• Spatial attentive deep learning (SADL) effectively performs automated placental segmentation on MRI scans.
• The SADL method shows high performance across two different gestational ages, supporting its utility in clinical practice.