Qualitative JPEG 2000 compression in digital mammography - evaluation using 480 mammograms of the CDMAM phantom

The DICOM standard supports both quantitative and qualitative lossy compression of mammograms.The purpose of this study was to investigate qualitative JPEG 2000 lossy compression and how different factors such as object thickness, radiation dose, and lossy compression levels affect image quality.

The CDMAM phantom Artinis 3.4 was radiographed with 4 different object thicknesses and 5 different doses. The images were compressed at 10 different compression levels. The image quality was assessed by the software interpolated IQFinv value.

Lossy 90 resulted in 89 % data reduction, lossy 70 in 95 % data reduction and lossy 60 in 96 % data reduction. At higher compression levels (lossy 30), the resulting image quality ranged from 80 - 36 %, and at low compression levels (lossy 90), it ranged from 89 - 93 %. The object thickness was found to significantly interact with the compression level with regard to the resulting image quality: a higher object thickness resulted in increasingly poor image quality at increasing compression levels (p < 0.05).

Higher qualitative JPEG 2000 compression levels contribute only little additional data reduction, while the resulting image quality cannot be reliably predicted. Factors affecting image quality such as radiation dose and object thickness should be taken into account when performing image compression. Large object thicknesses should be compressed with caution because the loss of image quality is greater when intelligent data compression algorithms are used.