research : projects: Medical Image Compression

Medical Image Compression project:


  • Create a professional peer-reviewed standard for medical image compression to which radiologists can refer when they want to improve storage and communication capabilities.

Value Proposition

  • Even if the cost of storage is dropping, the savings are surpassed by the increasing amount of data and the cost of operation remains high.
  • Access to high bandwidth remains limited and EHR networks cannot support large medical images.


Previous Study

  • Our previous pan-Canadian study allowed us to define average values for CR/DR, US, CT, MR and NM.
  • But we found significant degradation for CT images with JPEG2000 which triggered a subsequent study funded by Infoway to assess the effect of lossy compression on thin slices CT.
    The Effect of Compression on Subjective Assessment of CT Body Images
    The Effect of Compression on Subjective Assessment of Neurological Images
    J2K vs JPEG - Images compressed 15:1
    J2K compression ratio 15:1 JPEG compression ratio 15:1
    Noticeable degradation No noticeable degradation
    Difference in performance is due to the basis functions used to express the high frequency information in the white matter
  • DCT vs. DWT in CT Neuro
    • DCT: Good for compressing speckle pattern:
      • No spatial resolution
      • Similar objects (i.e. speckles) contribute to common DCT coefficients
      • Contribution from all speckles are added together which produces larger energy high-frequency DCT coefficients
      • Large energy coefficients are not discarded by quantization
    • DWT: Not good for compressing speckle pattern:
      • Has good spatial resolution
      • Few speckles are represented by few low energy DWT coefficient
      • Low energy high-frequency coefficients are discarded during quantization
      • Therefore, speckle pattern appears blurred

* Speckle pattern has significant high-frequency energy

  • Study finalized on 2008.
  • Results used in the CAR Standards for Irreversible Compression in Digital Diagnostic Imaging within Radiology, published on June 2008.
  • Results were also published on the Journal of Digital Imaging, Vol 22, No 6 (December), 2009: pp 569-578.


Current Study

  • JPEG and JPEG 2000 at 6, 9 and 12:1
  • Slices < 3 mm
  • 4 body parts: neuro, chest, body, MSK
  • up to 7 readers per session
  • Same methodology than previous study:
    • Diagnostic accuracy
    • Comparison side-by-side
  • The study evaluation and result analysis phases were finalized in December 2009.
  • Final Report and CAR Guidelines update have been submitted to the Canadian Association of Radiologists (CAR) for review & approval.