Every year, terabytes of data are generated through advanced imaging modalities such as MRI. Picture Archival and Communication Systems (PACS) are of immense interest to the medical community. The development of PACS depends critically on efficient compression algorithms for medical images.
One fast, memory-efficient way of processing volumetric medical images is through slice-wise predictive signal processing. In particular, for compression, this would lead to the usage of the block-matching algorithm in between MRI slices. However, this method proved to be very unsuccessful. In fact, compression with inter-frame estimation using this method proved to be worse than compressing frames individually, a very counter-intuitive result. This led to the notion that not much is to be gained by joint coding of MRI.
Using an improved interframe model, and in-loop noise filtering, we succeeded in achieving significant coding gain over single-frame coding of MRI (and all previous results). Thus, by overturning the common notion that joint coding of MRI frames is inefficient, this work has opened a slew of new possibilities in the area of medical image compression. In particular, others have followed with 3-D compression techniques that improve on our results at the cost of computational complexity.