Evaluation of image registration algorithms for 3D phase-resolved functional lung ventilation magnetic resonance imaging in healthy volunteers and chronic obstructive pulmonary disease patients.

Abstract

The purpose of the current study was to assess the influence of the registration algorithms on the repeatability of three-dimensional (3D) phase-resolved functional lung (PREFUL) ventilation magnetic resonance imaging (MRI). Twenty-three healthy volunteers and 10 patients with chronic obstructive pulmonary disease (COPD) underwent 3D PREFUL MRI during tidal breathing. The registration of dynamically acquired data to a fixed image was executed using single-step, stepwise, and group-oriented registration (GOREG) approaches. Advanced Normalization Tools (ANTs) and the Forsberg image-registration package were used for the registration. Image registration algorithms were tested for differences and evaluated by the repeatability analysis of ventilation parameters using coefficient of variation (CoV), intraclass-correlation coefficient, Bland-Altman plots, and correlation to spirometry. Also, the registration time and image quality were computed for all registration approaches. Very strong to strong correlations (r range: 0.917-0.999) were observed between ventilation parameters derived using various registration approaches. Median CoV values of the cross-correlation (CC) parameter were significantly lower (all p ≤ 0.0054) for ANTs GOREG compared with single-step and stepwise ANTs registration. The majority of comparisons between COPD patients and age-matched healthy volunteers showed agreement among the registration approaches. The repeatability of regional ventilation (RVent)-based ventilation defect percentage (VDPRVent ) and VDPCC was significantly higher (both p ≤ 0.0054) for Forsberg GOREG compared with ANTs GOREG. All 3D PREFUL-derived ventilation parameters correlated with forced expiratory volume in 1 s (FEV1 ) and the FEV1 / forced vital capacity (FVC) ratio (all |r| > 0.40, all p< 0.03). The image sharpness of RVent maps was statistically elevated (all p< 0.001) using GOREG compared with single-step and stepwise registration approaches using ANTs. The best computational performance was achieved with Forsberg GOREG. The GOREG scheme improves the repeatability and image quality of dynamic 3D PREFUL ventilation parameters. Registration time can be ~10-fold reduced to 9min using the Forsberg method with equal or even improved repeatability and comparable PREFUL ventilation results compared with the ANTs method.