Measurement Accuracy and Reproducibility of Semiautomated Metric and Volumetric Lymph Node Analysis in MDCT

作者：Buerke Boris^*; Puesken Michael; Mueter Stephan; Weckesser Matthias; Gerss Joachim; Heindel Walter; Wessling Johannes

来源：American Journal of Roentgenology, 2010, 195(4): 979-985.

DOI：10.2214/AJR.10.4010

• lymph nodes
• reproducibility
• segmentation
• volumetry

OBJECTIVE. The purpose of this study was to assess the measurement accuracy and reproducibility of semiautomated metric and volumetric lymph node analysis in MDCT.
MATERIALS AND METHODS. Whole-body CT with IV contrast administration was performed on 112 patients. Peripheral (cervical, axillary, and inguinal), abdominal, and thoracic lymph nodes were evaluated independently by two radiologists both manually and with semiautomated segmentation software. Long-axis diameter, short-axis diameter, and volume were measured. Agreement between the semiautomated and manual measurements (measurement error), need for manual correction, and relative interobserver differences were determined. Statistical analysis encompassed the variance inhomogeneity test, intraclass correlation coefficients, and Bland-Altman plots.
RESULTS. In total, 742 peripheral (cervical, axillary, and inguinal), abdominal, and thoracic lymph nodes (mean diameter, 13.2 +/- 4.3 mm; range, 4-37 mm) were evaluated. Semiautomatic segmentation without need for further correction was possible for 480 of 742 lymph nodes (64.7%). Calculation of intraclass correlation coefficients revealed high correlation between manual and semiautomatic measurements (r = 0.70-0.81) with a slight trend toward size overestimation for semiautomatic short-axis diameter (14.3%; limits of agreement, -34.3%, 62.9%) and long-axis diameter (11.7%; limits of agreement, -25.2%, 48.5%). Bland-Altman plots showed significantly (p < 0.0001) lower interobserver differences for semiautomated short-axis diameter (1.2%; 95% CI, -39.9% to 42.3%) compared with the manual measurement (7.6%; 95% CI, -38.7% to 53.9%). Among all locations, the relative interobserver difference for semiautomatic volume (2.9%; 95% CI, -31.4% to 37.3%) was significantly lower than that for manual short-axis diameter (p < 0.0001), manual long-axis diameter (0.0178), and semiautomatic short-axis diameter (p < 0.0001).
CONCLUSION. Semiautomatic short-axis diameter, particularly volume measurements, of lymph nodes are, irrespective of location, precise in terms of reproducibility and appear to be considerably more reliable than manual lymph node assessment.

• 出版日期2010-10