In Silico Metrological Assessment of 2D ROI Sampling Error Versus Whole-Organ 3D Volumetric CT Attenuation in Hepatic Steatosis: A Multicenter Monte Carlo Simulation on 971 Scans
- 1Faculty of Medicine, Badji Mokhtar University of Annaba
Abstract
PurposeTo quantify the metrological sampling error of conventional 2D regions of interest (ROIs) relative to whole-organ 3D volumetric CT attenuation ground truth in opportunistic screening for hepatic steatosis, and to evaluate its functional dependency on intraparenchymal spatial heterogeneity.
Materials and MethodsIn this retrospective computational investigation adhering to CLAIM and TRIPOD-AI reporting standards, 971 non-contrast and contrast-enhanced abdominal CT examinations were analyzed across a multicenter derivation cohort (TotalSegmentator v2, n = 841) and an external validation cohort (Medical Segmentation Decathlon Task03 Liver, n = 130). Automated 3D liver segmentation was conducted using nnU-Net, supplemented by an 8-mm spherical 3D morphological erosion to eliminate peripheral partial volume effects, liver capsule artifacts, and major hilar vascular structures. Three clinical sampling paradigms (1, 3, and 5 axial circular 2D ROIs of 15 mm diameter) were simulated through 500 stochastic Monte Carlo iterations per examination (485,500 total simulations). Absolute metrological sampling error (|{Delta}HU| = |HU_ROI - HU_3D|) and diagnostic steatosis misclassification rates at the standard unenhanced threshold (HU_3D [≤] 40 HU) were evaluated. Statistical significance across sampling paradigms was tested using two-tailed paired Wilcoxon signed-rank tests.
ResultsIn the derivation cohort (n = 841), the conventional guideline-recommended 3-ROI paradigm yielded a mean absolute sampling error of 6.61 {+/-} 7.83 HU (median: 2.92 HU; interquartile range: 1.34-8.45 HU). While representing a 41.6% error reduction compared to the single-ROI approach (11.32 {+/-} 14.16 HU; W = 0, p < 0.001), it demonstrated significant residual error compared to the extended 5-ROI approach (5.21 {+/-} 6.00 HU; W = 8, p < 0.001). Intraparenchymal spatial heterogeneity ({sigma}_HU) strongly predicted sampling error (Pearson r = 0.850, 95% CI: 0.830-0.868, p < 0.001). In borderline attenuation zones (35-45 HU), diagnostic misclassification reached 45.2% in heterogeneous livers. Findings were replicated in the external validation cohort (mean |{Delta}HU| = 7.38 HU for 1 ROI vs. 5.14 HU for 3 ROIs, p < 0.001).
ConclusionManual 2D ROI sampling exhibits an inherent physical uncertainty that exceeds the 5-HU clinical tolerance threshold in parenchymally heterogeneous livers, misclassifying nearly half of borderline examinations. Accurate opportunistic steatosis screening requires transition to automated whole-organ 3D volumetric CT densitometry embedded in PACS networks.
KEY POINTSO_LIConventional 3-ROI sampling leaves an average metrological error of 6.61 {+/-} 7.83 HU compared with 3D whole-liver ground truth. C_LIO_LIIntraparenchymal spatial heterogeneity is the primary determinant of 2D sampling error (r = 0.850, p < 0.001). C_LIO_LIIn borderline diagnostic attenuation ranges (35-45 HU), 2D manual sampling misclassifies up to 45.2% of patients with spatial heterogeneity. C_LI
Related
- WFUMB-Recommended Versus Manufacturer-Optimized Protocols for Ultrasound-Guided Attenuation Parameter in Hepatic Steatosis Assessment: A Retrospective Single-Vendor Multicenter Study
- Volumetric analysis and MRI radiomics for response assessment in canine glioma immunotherapy: an exploratory feasibility study
- Prevalence and Clinical Correlates of Coronary Perivascular Inflammation in an Asymptomatic Adult Primary Prevention Population: An Assessment of the CT-Derived Fat Attenuation Index Score (FAI-Score)
- Assessment of hepatic steatosis by controlled attenuation parameter using the M and XL probes: an individual patient data meta-analysis
- Visual rating versus volumetric segmentation of white matter hyperintensities for cognitive assessment: a memory clinic-based comparison and systematic review.
BibTeX
@misc{zemmour2026silico,
title = {{In Silico Metrological Assessment of 2D ROI Sampling Error Versus Whole-Organ 3D Volumetric CT Attenuation in Hepatic Steatosis: A Multicenter Monte Carlo Simulation on 971 Scans}},
author = {F. Zemmour},
howpublished = {medRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.10.26364608},
url = {https://www.medrxiv.org/content/10.64898/2026.10.10.26364608v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - In Silico Metrological Assessment of 2D ROI Sampling Error Versus Whole-Organ 3D Volumetric CT Attenuation in Hepatic Steatosis: A Multicenter Monte Carlo Simulation on 971 Scans
AU - F. Zemmour
PY - 2026
DA - 2026/10/10
PB - medRxiv
DO - 10.64898/2026.10.10.26364608
UR - https://www.medrxiv.org/content/10.64898/2026.10.10.26364608v1
AB - PurposeTo quantify the metrological sampling error of conventional 2D regions of interest (ROIs) relative to whole-organ 3D volumetric CT attenuation ground truth in opportunistic screening for hepatic steatosis, and to evaluate its functional dependency on intraparenchymal spatial heterogeneity. Materials and MethodsIn this retrospective computational investigation adhering to CLAIM and TRIPOD-AI reporting standards, 971 non-contrast and contrast-enhanced abdominal CT examinations were analyzed across a multicenter derivation cohort (TotalSegmentator v2, n = 841) and an external validation cohort (Medical Segmentation Decathlon Task03 Liver, n = 130). Automated 3D liver segmentation was conducted using nnU-Net, supplemented by an 8-mm spherical 3D morphological erosion to eliminate peripheral partial volume effects, liver capsule artifacts, and major hilar vascular structures. Three clinical sampling paradigms (1, 3, and 5 axial circular 2D ROIs of 15 mm diameter) were simulated through 500 stochastic Monte Carlo iterations per examination (485,500 total simulations). Absolute metrological sampling error (|{Delta}HU| = |HU_ROI - HU_3D|) and diagnostic steatosis misclassification rates at the standard unenhanced threshold (HU_3D [≤] 40 HU) were evaluated. Statistical significance across sampling paradigms was tested using two-tailed paired Wilcoxon signed-rank tests. ResultsIn the derivation cohort (n = 841), the conventional guideline-recommended 3-ROI paradigm yielded a mean absolute sampling error of 6.61 {+/-} 7.83 HU (median: 2.92 HU; interquartile range: 1.34-8.45 HU). While representing a 41.6% error reduction compared to the single-ROI approach (11.32 {+/-} 14.16 HU; W = 0, p < 0.001), it demonstrated significant residual error compared to the extended 5-ROI approach (5.21 {+/-} 6.00 HU; W = 8, p < 0.001). Intraparenchymal spatial heterogeneity ({sigma}_HU) strongly predicted sampling error (Pearson r = 0.850, 95% CI: 0.830-0.868, p < 0.001). In borderline attenuation zones (35-45 HU), diagnostic misclassification reached 45.2% in heterogeneous livers. Findings were replicated in the external validation cohort (mean |{Delta}HU| = 7.38 HU for 1 ROI vs. 5.14 HU for 3 ROIs, p < 0.001). ConclusionManual 2D ROI sampling exhibits an inherent physical uncertainty that exceeds the 5-HU clinical tolerance threshold in parenchymally heterogeneous livers, misclassifying nearly half of borderline examinations. Accurate opportunistic steatosis screening requires transition to automated whole-organ 3D volumetric CT densitometry embedded in PACS networks. KEY POINTSO_LIConventional 3-ROI sampling leaves an average metrological error of 6.61 {+/-} 7.83 HU compared with 3D whole-liver ground truth. C_LIO_LIIntraparenchymal spatial heterogeneity is the primary determinant of 2D sampling error (r = 0.850, p < 0.001). C_LIO_LIIn borderline diagnostic attenuation ranges (35-45 HU), 2D manual sampling misclassifies up to 45.2% of patients with spatial heterogeneity. C_LI
ER -