{
  "measurement": "Unchanged binary masks on their existing common arterial-ROI grid; no resampling or voxel repair.",
  "connectivity": "6-neighbour connected components, including every small component.",
  "parent_connection": "Raw source volumes are retained. For renal-branch connectivity only, both common TS parent trunks (aorta union IVC) are excluded from every system mask. Remaining vessel components 6-neighbour touching the respective common aorta/IVC are rooted; fraction is rooted branch voxels / all branch voxels. No inserted bridges. Branch metrics are NA unless both common parent masks exist.",
  "kidney_reach": "Number of common TotalSegmentator whole-kidney masks within <=3 mm Euclidean physical distance of rooted vessel voxels; an engineering proximity measure, not hilar entrances or vessel patency.",
  "bone_proxy": "Overlap with the common TotalSegmentator vertebrae/ribs/sacrum/hips/sternum prediction; an algorithmic proxy, not ground truth; also used by V8 and therefore not independent validation.",
  "agreement": "MONAI-V8 vessel-union Dice is inter-system agreement, not accuracy. Both raw-source and common-parent-excluded branch Dice are supplied. Whole kidney and renal parenchyma are different targets; no kidney Dice is calculated.",
  "surface": "FlyingEdges at 0.5 on the binary mask; full NIfTI affine to RAS millimetres; identical WindowedSinc smoothing with per-vertex displacement capped at 0.2 mm, no decimation/component removal/hole filling. Metrics precede smoothing.",
  "camera": "Shared CT-ROI physical bounds, orthographic projection; front camera at RAS +Y, back at -Y, superior +Z. Identical scale, colours, opacity, lights and surface processing for all systems within a case.",
  "timing": "No cross-system runtime table: TS parent-stage timing, MONAI inference and V8 postprocessing measure different scopes.",
  "monai": "Same public weights; first unshifted FP32 pass, sliding window overlap 0.5 and Gaussian blending. Displayed masks are independent sigmoid channels thresholded at 0.5 and may overlap. monai/merged_labels.nii.gz is separately retained as a mutually exclusive, lower-label-priority merge. Official bundle default overlap is 0.25. No TS parent vessels added to MONAI display; this adapted baseline is not a complete official workflow replication.",
  "v8": "Generic rule transfer using three translated-inference averages and general postprocessing. No new supervised training, no case-specific manual edits, and not equivalent to all previously edited private-case deliveries.",
  "urinary": "Only a confirmed descriptor E case receives a separate native-E candidate display. D is not treated as E; native-E masks are not silently fused/resampled into the A grid.",
  "clinical_boundary": "No manual reference standard, independence audit or clinical accuracy validation. Five cases are descriptive only; no sensitivity, accuracy, superiority claims, or significance testing.",
  "training_overlap": "Overlap of these public patients with any contributing model's training/validation data has not been excluded; this is not called independent external validation.",
  "phase_scope": "Five contrast-enhanced abdominal CT patients: only PUB-004 has explicit A/V/E; four have A/V/D delayed-phase extension inputs. They are not five CTU or five native E-domain validation cases. preprocessing_report.triphase_model_eligible only means registration quality passed, not that D satisfies the network's native E requirement. methods.json describes the actual run; protocol_review.md is a pre-run archive.",
  "totalsegmentator_scope": "The executed comparison is TotalSegmentator 2.14.0 total-family organ/parent-vessel tasks 291/293/297, without renal artery or renal vein branch labels. The newer official renal_arteries task 710 is present at commit c59be95b839252dd993e1e32ff3369168d9f4eab, requires a valid license, and has not been run or added to these frozen image columns. No dedicated renal-vein task was identified in the audited official task/class maps at that commit; this is a bounded source audit, not a claim that every official model lacks renal arteries."
}
