Lossless Compression of Brain Volumes

We implemented a lossless compression method tailored to brain volumes. By making use of the particular characteristics of these images, we sought to improve compression ratios compared to other general encoders. We exploit global and local symmetries of sub-bands after wavelet decomposition to achieve intra-band prediction. We also implement a shift-based inter-slice DPCM prediction scheme that exploits the correlation between slices of the volume [3]. After entropy-coding the final residuals of these stages using EBCOT, we compare our results against encoders including gzip, JPEG-LS, and JPEG-2000. In addition, the relative compression of brain volumes between normal subjects and multiple sclerosis patients was examined. Results showed that in general, our encoder performs on par with JPEG-LS and JPEG-2000 and compression ratios were lower in patients due to asymmetries in their brain structure.