PurposeTo develop and validate a standardized manual segmentation protocol for delineating choroidal tumors on B-scan ultrasonography to improve measurement reproducibility in uveal melanoma. DesignCross-sectional reproducibility study. Subjects, Participants, and/or Controls275 ultrasound images of melanocytic choroidal tumors obtained from a single tertiary ocular oncology clinic. MethodsTwo independent reviewers performed manual segmentation using an iterative, consensusbased annotation workflow, adjudicated and validated by an ocular oncologist.Inter-and intrareviewer agreement was assessed using surface-based metrics including average symmetric surface distance (ASSD), 95 th percentile Hausdorff distance (HD95), and normalized surface distance (NSD).The resulting layer segmentations were used to derive geometric tumor measurements including tumor height and largest basal diameter (LBD) measured using both a 1-chord and arclength approach. Main Outcome Measures J o u r n a l P r e -p r o o fInter-and intra-reviewer agreement (ASSD, HD95, NSD) and mean absolute error (MAE) for derived tumor thickness and LBD measurements. ResultsThe consensus protocol identified four consistent anatomical boundaries: inner retina, inner tumor margin, outer tumor margin, and outer sclera.Average inter-reviewer agreement across all classes was high, with ASSD: 0.14 mm, HD95: 0.49 mm, and NSD at 0.2 mm: 81%.Most disagreement occurred at the scleral interface and in cases with shadowing, artifact, subretinal fluid, or mushroom shape.Derived tumor measurements demonstrated low mean absolute errors of 0.21 mm, 0.59 mm, and 0.72 mm for thickness, 1-chord LBD, and arclength LBD, respectively. ConclusionsA standard, operationalized segmentation protocol enables reproducible boundary delineation on ophthalmic ultrasound and provides a foundation for consistent tumor measurement and future computational automation.
Chadwick et al. (Fri,) studied this question.