Evaluating reliability of automated quantitative brain morphometry from fetal T2-weighted MRI

Three-dimensional assessment of fetal cortical morphology from MRI is essential for understanding early brain neurodevelopment. However, measurement can be affected by fetal imaging quality, number and selection of available stacks, and reconstruction methods.

We evaluated the within-session reliability of an automated cortical morphometry pipeline in 30 typically developing fetuses [22-36 weeks gestational age (GA)]. For each subject, two disjoint subsets of 2D T2-weighted stacks (no shared stacks) were independently reconstructed into 3D volumes using the Neural Slice-to-Volume Reconstruction (NeSVoR) and the Slice-to-Volume Reconstruction Toolkit (SVRTK). Cortical plate volume, surface area, mean sulcal depth, and absolute mean curvature were extracted, and measurement reliability was assessed using absolute percent difference (APD) and intraclass correlation coefficients (ICC). Multiple linear regression evaluated the effects of mean stack quality, quality difference between subsets, stack count, and GA on measurement reliability.

NeSVoR-derived metrics showed high reliability for all measures (mean APD < 3%, ICC > 0.99). SVRTK-derived metrics were also robust (mean APD < 5%, ICC > 0.97). Reliability increased with greater stack count and older GA in NeSVoR, and with higher mean stack quality in SVRTK.

These results demonstrate that automated cortical morphometry from fetal MRI yields highly consistent measurements of volumetric and surface metrics within the proposed within-session design, once minimum levels of image quality and stack count are met. This study proposes a within-session benchmark for automated fetal cortical measurements and underscores that systematic reliability assessment is essential for confident application of automated pipelines in fetal neuroimaging.