Recent advances in cardiac MRI make real-time imaging of the heart possible. The combination of higher-performance gradient magnets and development of fast-scanning sequences allow image acquisition at rates of 6 frames per second, with sliding-window reconstruction techniques yielding video with rates up to 18 frames per second [1,2]. For 8-bit images of size 128x128, transmission of real-time video would require a channel that could support a bitstream of 2.3 Mbps.

The goal of this project is to implement a video compression algorithm with long-term memory motion-compensated prediction, and then investigate the bitrate savings when motion-compensated prediction and long-term memory motion-compensated prediction algorithms are applied to two test cardiac real-time studies. The first sequence, shown in Figure 1(a), is a short-axis view, or a slice through the left ventricle.

The sequence contrast causes the ventricle wall to appear dark, allowing the image reader to asses uniformity of wall motion. Motion is largely in-plane, so using displacement vectors to represent block motion might work better than for other sequences with through-plane motion and occlusion.

The second image sequence, Figure 1(b), is a view containing the right coronary. This sequence contains significant through-plane motion, making motion estimation from the previous image frame dificult. However, the sequence is periodic, so compression of this sequence using motion compensation with long-term memory prediction might reduce the error when compared to displacement vectors calculated from the previous frame.