Comparison of Manual, Semi-Automatic, and Automatic CT-Based Methods for Liver Volume Segmentation

Background/Objectives: To evaluate whether semi-automatic and automatic CT-based liver segmentation methods can provide clinically acceptable volumetric agreement compared with manual segmentation while improving processing efficiency in routine practice. Methods: CT images from 86 individuals were retrospectively analyzed. Liver volumes were calculated using manual segmentation, RVX Semi-Automatic, RVX Deep Learning, and TotalSegmentator. Differences among methods were assessed using repeated-measures ANOVA. Agreement with manual segmentation was evaluated using a Bland–Altman analysis, while the Dice Similarity Coefficient (DICE) and Hausdorff Distance (HD) quantified spatial overlap and boundary deviation, respectively. Processing times were recorded. Results: Mean liver volumes were 1503.9 ± 356.0 cm3 (manual), 1512.6 ± 373.6 cm3 (RVX Semi-Automatic), 1549.8 ± 367.9 cm3 (RVX Deep Learning), and 1518.3 ± 365.8 cm3 (TotalSegmentator). RVX Deep Learning produced significantly higher volumes compared with manual segmentation (p < 0.001), whereas RVX Semi-Automatic and TotalSegmentator showed no significant differences (p > 0.05). DICE values were 0.911 ± 0.032, 0.946 ± 0.018, and 0.938 ± 0.021 for RVX Semi-Automatic, RVX Deep Learning, and TotalSegmentator, respectively. HD values were highest for the deep learning-based method. Processing times were shortest for RVX Deep Learning and longest for manual segmentation. Conclusions: Semi-automatic and automatic liver segmentation methods substantially reduce processing time while maintaining clinically acceptable volumetric agreement. Among the evaluated approaches, TotalSegmentator showed the closest agreement with manual segmentation, supporting its use in routine CT-based liver volumetry. Deep learning-based segmentation, although faster, tended to overestimate volume, potentially limiting its use in applications requiring high volumetric precision.