Purpose: To compare lesion volume determination by applying diameter measurement and three different segmentation algorithms at different slice thicknesses reconstructed from computed tomography (CT) of a phantom model for hepatic colorectal metastases.
Materials and methods: Based on CT attenuation measurements obtained retrospectively from 20 patients with colorectal liver metastases, a phantom model was designed with a sponge soaked with a dilution of contrast agent and 6 embedded polyamide spheres (diameter, 8 - 30 mm) to simulate the contrast behavior of liver metastases. CT scans were obtained and reconstructed at different slice thicknesses (0.625/1.25/2.5/3.75 mm; increment, 1). One observer performed software-aided volume determination using the maximum diameter, manual segmentation, seed point method, and threshold method six times for each lesion in a randomized order. Statistical analysis revealed the absolute and relative differences from the actual lesion volumes and the intraobserver differences as well as the influence of slice thickness for each method.
Results: The mean relative differences of the seed point method (1.2 - 5.9%) and manual segmentation (2.6 - 4.9%) were significantly lower than the threshold method (5.4 - 12.8%) and diameter measurement (12.3 - 18.5%; p < 0.01). Volume determination by manual segmentation and the seed point method benefited from the use of thin-slice CT datasets. The intraobserver variation was lowest when using the manual segmentation (1.5 - 3.3%) and the seed point method (2.2 - 3.9%; p < 0.001).
Conclusion: Manual segmentation and the seed point method for thin CT slices were the methods with the lowest volume differences and intraobserver variation.
© Georg Thieme Verlag KG Stuttgart · New York.