RSNA 2012 

Abstract Archives of the RSNA, 2012


LL-INS-WE6B

Lossless Compression of Breast Tomosynthesis: Objects to Maximize DICOM Transmission Speed and Review Performance and Minimize Storage Space

Scientific Informal (Poster) Presentations

Presented on November 28, 2012
Presented as part of LL-INS-WE: Informatics Lunch Hour CME Posters  

Participants

David A. Clunie MBBS, Presenter: Officer, CoreLab Partners, Inc Owner, PixelMed Publishing LLC Research support, Siemens AG Advisory Board, CoreLab Partners, Inc

CONCLUSION

Standard DICOM compression schemes, especially JPEG-LS, should be used for breast tomosynthesis data.

BACKGROUND

Availability & use of digital breast tomosynthesis is increasing & the size of the reconstructed images are an infrastructure challenge. Proprietary solutions are feasible, but it is desirable to re-use PACS standards & infrastructure, including archives & viewers. DICOM has a standard object for this purpose, & standard compression transfer syntaxes to minimize data size.

DISCUSSION

The JPEG-LS scheme is not widely implemented in PACS at this time, though standardized in DICOM. It only slightly outperformed J2K in this experiment, but is very simple and much faster than J2K & much less demanding on memory & hardware performance; this speed benefit may be relevant in high volume screening applications, so consideration should be given to "encouraging" PACS vendors to implement it for this reason. Multi-component JPEG 2000 is also not widely implemented, though showed the best results. A multi-component form of JPEG-LS might be ideal, if it were to be standardized.The bzip2 non-image scheme performed remarkably well, and should be considered for inclusion in DICOM.

EVALUATION

Implementations of various standard DICOM lossless (reversible) compression transfer schemes were selected, including deflate, JPEG lossless, JPEG-LS, & JPEG 2000 (both single slice and multi-slice).De-identified samples of various vendors' tomosynthesis images were obtained & the pixel data extracted from the header. The size of compressed images was compared by overall file size, size of the bit stream without padding & against zero order entropy (amount of information present in the image, ther being lots of air & small breasts occupying varying amountds of the reconstructed area).Results show that all schemes provide at least modest compression (3:1 relative to file size), image-specific compression schemes out-perform generic compression schemes (except bzip2, which is not standardized in DICOM), JPEG lossless performs the worst of the image-specific schemes, & JPEG-LS performs as well or better than JPEG 2000 ( > 4 or 5:1), unless multi-slice (multi-component transform) JPEG 2000 is used.

Cite This Abstract

Clunie, D, Lossless Compression of Breast Tomosynthesis: Objects to Maximize DICOM Transmission Speed and Review Performance and Minimize Storage Space.  Radiological Society of North America 2012 Scientific Assembly and Annual Meeting, November 25 - November 30, 2012 ,Chicago IL. http://archive.rsna.org/2012/12037671.html