Semiautomatic lymph node segmentation in multislice computed tomography: impact of slice thickness on segmentation quality, measurement precision, and interobserver variability.

OBJECTIVES

To determine the impact of slice thickness on semiautomatic lymph node analysis.

MATERIALS AND METHODS

Thoracic multislice computed tomography (MSCT) of 46 patients with NSCLC were reconstructed at 1.0-, 3.0-, and 5.0-mm slice thickness. Two radiologists independently determined long and short axis diameter and volume of 299 thoracic lymph nodes by semiautomatic segmentation software. Necessity of manual correction (= relative difference between uncorrected and corrected segmented lymph node volume) and relative interobserver differences were determined. The precision of segmentation was expressed by relative measurement deviations (RMD) from the reference standard (mean of 1.0 mm datasets). Statistical analysis encompassed t test and Bland-Altman plots.

RESULTS

Necessity of manual correction was significantly higher for 5.0 mm than for 3.0 (P = 0.042) or 1.0 mm (P = 0.0012). The RMD for long and short axis diameter were found to be independent of slice thickness, whereas the RMD for lymph node volume significantly (P = 0.021) increased from 4.0% at 1.0 mm (95% CI: 1.0%-3.5%) to 35% at 5.0 mm (95% CI: 10.5%-60.5%). The relative interobserver differences was consistently low for metric and volumetric parameters (eg, volume 2.3%, 95% CI: -7.4%-10.8% at 5.0 mm) with no difference in any of the slice thicknesses (P > 0.064).

CONCLUSIONS

Significant deviations in lymph node volume together with excessive manual corrections suggest reconstruction of the data for semiautomatic lymph node assessment at a slice thickness of 1.0 mm but not exceeding 3.0 mm.