Liver lesion segmentation in MSCT: effect of slice thickness on segmentation quality, measurement precision and interobserver variability.

Abstract

To evaluate the effect of slice thickness on semi-automated liver lesion segmentation.
In this retrospective study, liver MSCT scans from 60 patients were reconstructed at a slice thickness of 1.5 mm, 3 mm and 5 mm. 106 liver lesions (8 - 64 mm, mean size 25 ± 13 mm) were evaluated independently by two radiologists using semi-automated segmentation software (OncoTreat®). Lesions were classified as cystic, hypodense and hyperdense according to their contrast-to-noise ratio (CNR). The long axis diameter (LAD), short axis diameter (SAD) and volume were measured. The necessity for manual correction (NOC = relative difference between uncorrected and corrected volume) and the relative interobserver difference (RID) were determined. Precision was calculated in terms of relative measurement deviations (RMD) from the reference standard (mean of 1.5 mm data sets). Wilcoxon test, t-test and intraclass correlation coefficients (ICC) were employed for statistical analysis. All statistical analyses were intended to be exploratory.
Regardless of the liver lesion subtype, the NOC was found to be significantly higher for 5 mm than for 3 mm (p = 0.035) and 1.5 mm (p = 0.0002). The RID was consistently low for metric and volumetric parameters with no difference in any of the slice thicknesses for all subtypes (ICC > 0.89). The RMD increased significantly for the LAD, SAD and volume at a slice thickness of 5 mm (p < 0.01), e. g. volume: 0.5 % at 1.5 mm, 5.5 % at 3.0 mm and 7.6 % at 5.0 mm.
Since the deviations in measurements are significant, and manual corrections made during semi-automated assessment of the liver lesions are considerable, a slice thickness of 1.5 mm, and no more than 3.0 mm, should be used for reconstruction for inconsistently vascularized liver lesions.