Segmentation repeatability and correlation of texture and shape features with gestational age in MRI of healthy placenta, placenta previa, and placenta accreta spectrum

Objectives

To determine the most repeatable anatomical plane for placental segmentation and investigate any correlation between gestational age (GA) and texture and shape features of healthy placenta, placenta previa (PP) and placenta accreta spectrum (PAS).

Methods

T2 MRIs from 190 women with PP, PAS and healthy placenta were collected from a single center. Intra- and inter-observer studies were conducted by 3 operators, each manually segmenting 3 times in the sagittal, coronal, and axial planes. Two thousand five hundred and twenty-five texture and shape features were extracted from each MRI using PyRadiomics, and their correlation with GA determined using Spearman’s correlation coefficient (⁠ ⁠).

Results

For the segmentation study, 3 healthy and 3 PAS pregnancies were investigated. The sagittal plane exhibited the highest median dice coefficient (DC) in the intra-observer study for PAS (0.915) and inter-observer study for healthy (0.907) and PAS (0.891). For correlation with GA, 46 healthy, 49 PAS, and 57 PP MRIs were utilized. Texture features like grey level non-uniformity exhibited strong correlations with GA in the healthy group (⁠ = 0.81, P < .001), but only moderate correlations in the PP (⁠ = 0.58, P < .001) and PAS (⁠ = 0.52, P < .001) groups.

Conclusion

The sagittal plane was the most repeatable anatomical plane for manually segmenting healthy placenta and PAS. Texture features describing heterogeneity were correlated with GA, particularly in healthy placenta.

Advances in knowledge

Texture and shape features captured the increase in heterogeneity seen throughout gestation on placental MRI, indicating a need to consider GA when investigating textural differences in healthy placentas, PP, and PAS.