Semiautomatic determination of arterial input functions for quantitative dynamic contrast-enhanced magnetic resonance imaging in non-small cell lung cancer patients.

Abstract

The aim of this study was to validate a semiautomatic detection method for the arterial input functions (AIFs) using Kendall coefficient of concordance (KCC) for quantitative analysis of dynamic contrast-enhanced magnetic resonance imaging in non-small cell lung cancer patients. We prospectively enrolled 28 patients (17 men, 11 women; mean age, 62 years) who had biopsy-proven non-small cell lung cancer. All enrolled patients underwent dynamic contrast-enhanced magnetic resonance imaging of the entire thorax. For the quantitative measurement of pharmacokinetic parameters, K and ve, of the lung cancers, AIFs were determined in 2 different ways: a manual method that involved 3 independent thoracic radiologists selecting a region of interest (ROI) within the aortic arch in the 2D coronal plane and a semiautomatic method that used in-house software to establish a KCC score, which provided a measure of similarity to typical AIF pattern. Three independent readers selected voxel clusters with high KCC scores calculated 3-dimensionally across planes in the data set. K and ve were correlated using intraclass correlation coefficients (ICCs), and Bland-Altman plots were used to examine agreement across methods and reproducibility within a method. Arterial input functions were determined using the data from ROI volumes that were significantly larger in the semiautomatic method (mean ± SD, 3360 ± 768 mm) than in the manual method (677 ± 380 mm) (P < 0.001). K showed very strong agreement (ICC, 0.927) and ve showed moderately strong agreement (ICC, 0.718) between the semiautomatic and manual methods. The reproducibility for K (ICCmanual, 0.813 and ICCsemiautomatic, 0.998; P < 0.001) and ve (ICCmanual, 0.455 and ICCsemiautomatic, 0.985, P < 0.001) was significantly better with the semiautomatic method than the manual method. We found semiautomated detection using KCC to be a robust method for determining the AIF. This method allows for larger ROIs specified in 3D across planes as opposed to manually selected ROIs restricted to the 2D coronal images seen by the reader and provides increased reproducibility that is comparable with manual specification.