Comparison of Manual, Semi- and Fully Automated Heart Segmentation for Assessing Global Left Ventricular Function in Multidetector Computed Tomography
C. Plumhans,S. Keil,C. Ocklenburg,G. Mühlenbruch,F. Behrendt,R. Günther,A. Mahnken
DOI: https://doi.org/10.1097/RLI.0b013e3181aaf4e1
IF: 6.7
2009-08-01
Investigative Radiology
Abstract:Purpose:To evaluate the reliability of global left ventricular (LV) function and mass measurements with the aid of a semi-automated (Circulation; Siemens, Forchheim, Germany) and a new fully automated software (Philips Research Europe, Aachen, Germany) versus an established manual segmentation method (Argus; Siemens). Material and Methods:Forty-one patients (31 men, 10 women; mean age: 62 ± 5 years) with known or suspected coronary heart disease underwent contrast-enhanced Dual-Source computed tomography of the heart (120 kV, 410 mAs/rotation, collimation 2 × 32 × 0.6 mm, gantry rotation time 0.33 milliseconds). Global LV function measurements of end-diastolic volume (EDV), end-systolic volume (ESV), stroke volume, ejection fraction (EF), and LV mass were each assessed with a manual, a semi- and fully automated method. The latter were compared with the manual contour tracing method, which was considered as standard of reference. Postprocessing time for each method was recorded. For statistical analysis, repeated-measures analysis of variance, post hoc t test, and concordance correlation coefficients were calculated. Bland-Altman plots were generated. Results:In general, ESV and EF assessed with the semi-automated and with the fully automated prototype version agreed well with the manual contour tracing method. The mean ESV (±SD) calculated from the manual, the semi-automated, and the fully automated method was 67 ± 43 mL, 74 ± 54 mL, and 75 ± 48 mL, respectively. No statistically significant differences between the methods were found for ESV and EF. In contrast, significant variations (P < 0.05) among the different segmentation methods were shown for EDV, stroke volume, and LV mass. This variation was predominantly due to variation in endocardial delineations among the different techniques. Concordance correlation coefficients demonstrated a better accuracy for the fully automated method than for the semi-automated technique when compared with the manual drawing method. Furthermore, fully automated postprocessing heart segmentation yielded time savings of approximately 80% compared with the manual segmentation tool and 63% compared with the semi-automated technique. Mean postprocessing time (±SD) for the manual, the semi-automated, and the fully automated method was 345 ± 75 seconds, 192 ± 58 seconds, and 72 ± 58 seconds, respectively. Conclusion:LV function and mass analyses using semi- or fully automated segmentation algorithms are feasible even if significant differences in EDV assessment are observed. The fully automated method results in better accuracy and time savings when compared with manual and semi-automated data analysis.