Fully automated registration and warping of contrast-enhanced first-pass perfusion images.

Respiratory motion during acquisition of first-pass myocardial perfusion images results in translation, distortion from out-of-plane motion, and changes in left ventricular geometry. Together these effects make visual image analysis more difficult and limit methods of quantitative analysis of contrast kinetics. We present a fully automated registration and warping algorithm for correcting translation and geometric distortions using a statistically based image registration method. Twelve patients (mean age 51 +/- 12 years) were studied 3 +/- 1 days after reperfused first myocardial infarction. Perfusion images were acquired during bolus administration of nonionic Gd-DTPA. Pixel intensity statistics were computed for each image in the neighborhood of high spatial frequencies. These statistics were then used to register and warp each target image (image to be registered and warped) to a common template image. Average image-to-image vertical translation was 2.6 +/- 0.8 pixels (3.4 +/- 1.0 mm) prior to processing and 0.9 +/- 0.3 pixels (1.2 +/- 0.4 mm) post-processing (P < 0.0001). Mean image-to-image horizontal translation was 1.7 +/- 1.2 pixels (1.8 +/- 1.2 mm) before and 1.3 +/- 0.7 pixels (1.4 +/- 0.7 mm) after processing (P = 0.05). Left ventricular endocardial area varied an average of 105 +/- 55 pixels (140.7 +/- 53.7 mm2) between images prior to processing vs. 51 +/- 15 pixels (68.3 +/- 20.1 mm2) after processing (P < 0.001). Thus automated, statistically based registration and warping of perfusion images is effective in reducing image-to-image translation. This method may permit more sensitive qualitative and quantitative evaluation of myocardial contrast-enhanced first-pass images.