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

Area of Science:

• Medical Imaging
• Radiology
• Computational Anatomy

Background:

• Accurate liver volumetry is crucial for clinical decisions.
• Manual segmentation is time-consuming and prone to variability.
• Automated methods aim to improve efficiency and consistency.

Purpose of the Study:

• To compare semi-automatic and automatic CT liver segmentation methods against manual segmentation.
• To evaluate volumetric agreement and processing efficiency.
• To identify optimal methods for routine clinical practice.

Main Methods:

• Retrospective analysis of CT images from 86 individuals.
• Liver volumes segmented manually and using RVX Semi-Automatic, RVX Deep Learning, and TotalSegmentator.
• Statistical analysis included ANOVA, Bland-Altman analysis, Dice Similarity Coefficient (DICE), and Hausdorff Distance (HD).

Main Results:

• RVX Deep Learning showed significantly higher volumes (p < 0.001); others had no significant difference.
• TotalSegmentator and RVX Semi-Automatic demonstrated high volumetric agreement with manual segmentation.
• RVX Deep Learning was fastest, manual segmentation was slowest; DICE scores were high for all automated methods.

Conclusions:

• Semi-automatic and automatic liver segmentation significantly reduce processing time.
• TotalSegmentator offers the closest agreement to manual segmentation for routine CT liver volumetry.
• Deep learning methods are faster but may overestimate volumes, impacting precision-sensitive applications.