ID #194 Development and validation of a standardized step-by-step manual segmentation protocol for hypothalamic volume in pediatric craniopharyngioma
Brianna Smith, Jacinto Carrasco, Maxine Kugelmas, Bailey Tanner, Leticia Sewaybricker, Greta Wilkening, Christina Chambers, Todd Hankinson, Allison ShapiroAbstract
Introduction
Adamantinomatous craniopharyngiomas (ACP) frequently cause hypothalamic injury, leading to significant endocrine, metabolic, and cognitive morbidity. Accurate hypothalamic volumetry may serve as a biomarker of postoperative injury, yet existing segmentation methods are not optimized for pediatric patients with distorted anatomy. We developed a standardized, reproducible manual segmentation protocol tailored to pediatric patients with ACP.
Methods
High-resolution T1-weighted MRI scans were preprocessed using FSL and upsampled to 0.4 mm³ isotropic resolution. Anatomical boundaries were defined using adapted neuroanatomical criteria. Two trained raters, blinded to clinical data, independently segmented the hypothalamus in ITK-SNAP with standardized contrast settings and overlay with a gray matter tissue probability map. Iterative protocol refinement was performed to improve interrater reliability. Cases with >20% volumetric discrepancy were reviewed by a third rater. All final segmentations were adjudicated by a pediatric neurosurgeon. Cases with > 5% discrepancy after neurosurgeon review were re-segmented.
Results
Across 20 participants, raters differed by an average of 1.45, 2.0, and 3.8 slices for the anterior, superior, and posterior boundaries. Volumetric estimates differed by < 20% in 13 cases; six required third-rater consensus review. After neurosurgical adjudication, one case required re-segmentation due to boundary inconsistencies > 5% difference. The median percent differences in boundary delineation after neurosurgical adjudication were 1.16% (IQR: 0.16–1.61%) for the anterior boundary and 0.37% (IQR: 0–0.66%) for the superior boundary. One participant was excluded due to severe anatomic distortion precluding hypothalamic identification.
Conclusions
This study presents the first standardized manual segmentation protocol for hypothalamic volumetry in pediatric ACP. The protocol demonstrated strong interrater reliability despite substantial anatomic variability and provides a reproducible framework for future studies. This approach supports the use of hypothalamic volume as a potential biomarker of postoperative injury and long-term morbidity in pediatric craniopharyngioma.