Automated transformation of dynamic cardiac images in the short axis view

Objectives Myocardial blood flow images (MBF) are typically acquired with the patient in the supine position. To facilitate clinical interpretation, the data are transformed manually to the short axis view. The purpose of this work is to eliminate the manual transformation, opening the way to fully automated quantitative blood flow imaging in a standardized anatomical reference frame.

Methods Cardiac dynamic 13N-NH3 PET images were acquired in 5 patients imaged in the supine position. Generalized factor analysis of dynamic sequences (GFADS) decomposed each voxel as a combination of three products, the factor images and factor curves. Previous experience shows that these factors can be identified as left ventricle, right ventricle and myocardium. A 12 parameter affine transformation was used to register the 2 ventricular factor images (blood pool) from each subject to reference factor images in the short axis view. The transformation matrix was then applied to the PET images to effect their transformation to the short axis view. The accuracy of the transformation was evaluated by comparison to the standard manual method. Additional analyses included higher quality images from 18F-Flurpiridaz data acquired in pig.

Results The short axis view images obtained with the proposed technique were similar to the manual method. Better results were obtained for 18F-Flurpiridaz data with a reproducibility (fractional s.d. of myocardium voxel values across replicates) of 2% for our method and 5% for the manual method.

Conclusions GFADS with affine registration can be used to transform dynamic cardiac images to a short axis template without operator intervention. Registration based on blood pools rather than tracer uptake in myocardium should be less sensitive to myocardial pathologies and the use of affine transformation with scaling and shearing should yield better accuracy than simple translation and rotation as in the manual method. Our method may allow extension of the bull's eye approach allowing automated evaluation of MBF in clinical and research environments.