Development of a Heart Motion Tracking System using Non-invasive Imaging Data

Abstract

Cardiac motion can be monitored non-invasively for the assessment of cardiovascular function by using medical imaging systems and motion tracking algorithms. Existing tracking approaches require a priori understanding of the non-rigid motion of the target system, which could change over multiple cardiac cycles and lead to tracking failures. The purpose of this research is to develop the algorithm and software, with computer vision techniques, to continuously track the motion of a user-defined region of the heart images. The proposed algorithm improves upon existing techniques because it does not require an underlying motion model, it quantifies the quality of tracking, and it can recover from a failed tracking estimate. The motion estimation of a non-rigid system will be done by a piecewise tracking approach that breaks up the region of interest into several small segments (patches), which can be approximated with interconnected pseudo-rigid segments. These segments will be initialized based on two criteria: 1) motion within a segment must follow the pseudo-rigid body model; and 2) motion in neighboring segments must be similar to each other. Segments are subsequently tracked as pseudo-rigid bodies, and the criteria described above are also used to detect failures in tracking. If a failure were to occur, the tracking algorithm will be reinitialized automatically. This algorithm was shown to be accurate and efficient, and has been tested on several heart motion data sets.