Semi-automated CT attenuation measurement for MASLD: volumetric versus peripheral-ROI methods

This study aimed to evaluate semi-automated liver CT attenuation measurement methods for assessing metabolic dysfunction-associated steatotic liver disease (MASLD) severity. Adult patients suspected of having MASLD were retrospectively recruited from Zhujiang Hospital between June 2020 and June 2022. Liver segmentation was performed using nnU-Net, with manual refinement performed in 71 of 179 cases (39.7%). Two distinct semi-automated quantification approaches: (1) whole-liver volumetric analysis and (2) peripheral region-of-interest (ROI) measurement, were implemented to objectively assess hepatic steatosis severity while eliminating manual measurement variability. The associations between steatosis level and MRI-PDFF values were analyzed using Spearman’s correlation coefficient. The diagnostic performance of hepatic steatosis severity (mild: MRI-PDFF ≥ 5%; moderate: MRI-PDFF ≥ 14%; severe: MRI-PDFF ≥ 28%) was evaluated through receiver operating characteristic (ROC) curve analysis and compared to manual measurement outcomes. A total of 179 patients were included (mean age, 34.38 years ± 10.67, 96 men), with 150 exhibiting hepatic steatosis (MRI-PDFF ≥ 5%). The peripheral-ROI method outperformed volumetric analysis for detecting hepatic steatosis with MRI-PDFF ≥ 5%, achieving AUCs of 0.809 (95% CI: 0.677, 0.941; P < .001), 0.854 (95% CI: 0.776, 0.931; P < .001), and 0.655 (95% CI: 0.454, 0.856; P < .001) for the ≥ 5%, ≥ 14%, and ≥ 28% thresholds, respectively. Compared with manual measurement, the peripheral-ROI method showed comparable performance at the ≥ 5% and ≥ 14% thresholds but was significantly inferior at the ≥ 28% threshold ( P = .041). The peripheral-ROI measurement technique based on nnU-Net automatic segmentation achieves classification performance comparable to manual measurement for mild-to-moderate steatosis (MRI-PDFF < 28%), while manual measurement remains superior for severe steatosis (MRI-PDFF ≥ 28%). Longitudinal observation suggests its potential for follow-up monitoring in clinical practice, though further studies with larger cohorts are needed.