Anatomy-prior-guided and scan-aware 3D segmentation of the pediatric pancreas in CT images.

OBJECTIVE: Pediatric pancreas CT segmentation is challenging because developmentalanatomy and scan-protocol variability jointly affect organ size, image appearance, andboundary visibility. This study evaluated whether explicitly incorporating anatomy-relatedand scan-related case-level descriptors could improve pediatric pancreas segmentationrelative to an image-only baseline. APPROACH: We developed APSC-Net, ananatomy-prior-guided and scan-aware 3D segmentation framework based on SegResNet.Structured anatomy-related and scan-related descriptors were encoded as separate contextvectors and injected through bottleneck anatomy-prior guidance and decoder-stagelow-rank FiLM modulation. The model was trained and internally tested on GE pediatricCT cases, then evaluated without retraining on a held-out Siemens subset in across-vendor evaluation within the same cohort and institution. MAIN RESULTS: On the fullheld-out GE test set, APSC-Net improved the mean Dice similarity coefficient (DSC) from0.665 to 0.770 and reduced the 95% Hausdorff distance (HD95) from 22.2 to 11.6 mmcompared with an image-only SegResNet baseline. Stratified analysis showed DSC gainsacross pediatric age groups, including the youngest subgroup (0.718 to 0.751 for 0-5years), and across exposure-stage strata. In the GE stream ablation, the full and scan-onlyconfigurations performed similarly. In secondary testing on held-out Siemens cases, the fullmodel achieved lower ASSD than either single-stream variant and achieved the lowestHD95 and ASSD among the compared methods, while age-group domain-regularizedbaselines remained competitive in DSC. SIGNIFICANCE: On the evaluated cohorts, APSC-Netachieved lower boundary-distance errors than the image-only SegResNet baseline. Thefindings support further evaluation of case-level metadata conditioning for pediatricimaging tasks affected by developmental and acquisition-related variability.