Motion Estimation in Cardiac Fluorescence Imaging

Motion artifacts are a major disadvantage of cardiac optical mapping studies. Pixel misalignment due to contraction is a main cause of the presence of gross motion artifacts in action potential recordings. This study is focused on methods for identifying landmarks and tracking the motion of cardiac tissue for preparations in optical mapping recordings. This is a first step toward our long-term goal to implement a landmark-based image registration technique to correct for pixel misalignment in cardiac optical mapping fluorescence videos and, hence, for gross motion artifacts. Preliminary results for the registration step are presented as an initial proof of concept. The characteristics of the optical mapping images are challenging, since their lack of contrast and well-defined features impose a limitation on the techniques than can be used for landmark selection and motion tracking.

This paper compares results of motion estimation of the cardiac surface with two approaches that do not rely on high-contrast features: 1) Scale-invariant feature transform (SIFT) detected “keypoints,” to be used as landmarks for motion tracking, as well as 2) a classical global optical flow (OF) algorithm. Both are applied to low-contrast and low-resolution cardiac fluorescence images. We demonstrate that the performance of SIFT is superior to that of OF for pixel motion tracking in cardiac optical mapping images with simulated motion. Results for action potential recovery and action potential duration calculation after landmark-based image registration show that SIFT landmark-based registration yields superior performance in this regard as well.