Modern PACS solutions such as JiveX are already able to configure the JPEG 2000 algorithm and thus automatise the compressionrates that were suggested by different medical societies in several countries (Canada, Germany, and UK) and recently confirmed by the ESR position paper. The information on modality and anatomy which is required to control compression is read from the headers of the DICOM data and compared to the defined rules. This allows lossy compression of image data according to the preset values.
However, as Dr. Marc Kämmerer, JiveX product manager and radiologist by training, concedes, "users still hesitate to use this functionality" since they cannot be sure that an image after lossy compression still contains all diagnosis-relevant information. A reliable automatized quality control procedure does not yet exist - this is an important task for the industry to tackle.
"As far as the technical realisation of automatized quality control is concerned, we are still on square one. Before we move on we have to answer some basic questions such as: Does compression affect all image parts, even the informal ones? Or: Do the grayscale values correspond to the real values contained in the image?", explains Dr. Kämmerer. Mathematically speaking there are different approaches for individual parameters such as pixel-to-noise ratio (PNR), Just Noticeable Difference (JND) or Mean Square Error. By themselves these parameters, however, do not provide sufficient information whether the lossy compressed image still contains all diagnosis-relevant information. In view of the fact that such a virtual quality control is burdened with so many variables and unknowns joint efforts are required to come up with a solution. "This is why we launched the ESR initiative which brought together experts from different disciplines and from different countries. Together we want to find out whether the individual parameters combined with further abstract descriptions will allow us to create a decision making matrix. Such a matrix should be a tool to classify the diagnostic content of an image after compression. What we already do know, however, is that such a matrix of diagnostic validity of an image most likely will not provide a simple digital yes/no decision. Rather we expect confidence intervals which asses the probability of diagnostic validity," Kämmerer explains. Firstly, extensive research is required to establish the salient aspects of such a matrix. That means modalities and image settings have to be tested and analysed, or as Kämmerer puts it: "This is a gigantic task which needs to be performed as a multi-centre study at large universities. One institution has to manage that study and this is where ESR might come in."
It will take a few years for a quality-control solution to take shape. Until then, however, the data compression capabilities offered for example by JiveX need not lie idle, Marc Kämmerer underlines: "In Germany compression according to the results of the German consensus conference is accepted by most public bodies, for example those that are in charge of authorizing teleradiology procedures pursuant to the x-ray ordinance. This should reduce the radiologists´ uncertainty. Nevertheless, the decision in favour of lossy compression will remain to a large extent an individual one. If lossy compression is to be used in primary diagnostics, samples are required to confirm the integrity of diagnostic information.
Quality control of lossy data compression in the context of long-term archiving is another interesting issue. The x-ray ordinance explicitly allows data compression but requires maintaining diagnosis-relevant information. Over the next few years, we as PACS providers will no doubt invest many resources in refining the technology in order to make lossy data compression with automatized quality control an easy-to-use and safe tool for all users."
Article: Usability of irreversible image compression in radiological imaging. A position paper by the European Society of Radiology (ESR): Download PDF