Introduction | MCP | Matlab Investigation | C Implementation | Results | Conclusions | References | Appendix

Results

We showed the original cardiac sequences (full 18 frames per second) as well as the reconstructed sequences at various PSNR levels to a cardiologist who works with real-time cardiac MRI studies. Surprisingly, he indicated that the diagnostic utility of these particular sequence types was preserved at a PSNR of between 30 and 31 dB. We had previously supposed, based on tolerance for lossy compression in other types of static MR images with which we are more familiar, that this kind of PSNR would yield images of unnacceptable quality. At this PSNR, our results showed very little difference between long-term memory MCP, simple MCP, and frame-difference encoding without MCP. The compression ratio obtained by our encoder at this PSNR is roughly 32:1. This yields a bit rate of approximately 70 kbps for the 18 frame/sec studies.

Shown below are the original vs. reconstructed sequences at a PSNR of approximately 31 dB, which, as indicated, still preserve their diagnostic utility. As can be seen, there is very little perceptible difference between the original and reconstructed sequences even at this PSNR.

a) b)
Figure 2. a)
Original short-axis view. b) Compressed using long-term memory MCP and most coarse quantization size.

a) b)
Figure 2. a)
Original short-axis view. b) Compressed using long-term memory MCP and most coarse quantization size.



Introduction | MCP | Matlab Investigation | C Implementation | Results | Conclusions | References | Appendix