Before displaying an ultrasound image, the machine usually performs a logarithmic compression in order to better visualise an image in terms of grayscale. In order to treat a DICOM image, is it necessary to do a logarithmic decompression for more accurate results of image analysis and processing?
It depends on the type of processing and analysis you want to perform. If the information you are looking for is encompassed in pixel values, if you do not inverse the logarithmic compression, the values that you will extract will depends on the compression and its parameter, for instance the dynamic range that has been used. This is problematic if you want to be able to compare results for acquisitions with different parameters. Moreover, if the information you are looking for is proportional to the linear echo-power, you also want to get rid of logarithmic compression.
However you have also to keep in mind that 1) generally the true compression equation is not provided by the manufacturers 2) the parameters displayed on the screen may not be right 3) After decompression, your data are still affected by effects such as tresholding or quantization.
We have tackled this issue in our publication of IEEE UFFC, 2005 on despeckle filtering analysis of ultrasound images:
C.P. Loizou, C.S. Pattichis, C.I. Christodoulou, R.S.H. Istepanian, M. Pantziaris, A. Nicolaides “Comparative evaluation of despeckle filtering in ultrasound imaging of the carotid artery,” IEEE Trans. Ultras. Ferroel. Freq. Contr., vol. 52, no. 10, pp. 1653-1669, 2005.
as well as in some other publications made by our group.
More specifically we show that Logarithmic compression is applied to the
envelope-detected echo signal in order to fit it in the display range. This compression affects the speckle noise statistics and the high-intensity tail of the Rayleigh and Rician probability density function. As a result, the speckle noise becomes very close to white Gaussian noise corresponding to the uncompressed Rayleigh signal. Because the effect of additive noise is considerably smaller compared with that of multiplicative noise it can be left out.
In any case logarithmic decompresion is not necessary. As long as all processing occurs on the compressse signal than results are comparable.
Unless you know exactly what kind of compression has been carried out on the image, it is NOT advisable to attempt any decompression, as it might degrade the image. You can directly work with the image.
If you are looking to characterize physical quantities, for example the perfusion of tissue using contrast enhanced ultrasound, then you need to go to linear data to be able to perform the modelling of the haemodynamics. Otherwise you wouldn't be able to relate the intensity signal with the concentration of the contrast agent.
As Barrois mentioned, even if you perform careful measurements at the settings that you are imaging to map your pixel intensity values to linear intensity you still get a quantization error that unfortunately increases with intensity due to the shape of the curve. (e.g. a small increment in pixel value results in a large increment in linear intensity). Furthermore you have saturation even with linear data.
Unfortunately most of the times only the manufacturer has access to the linear intensity values.
By the way the DICOM images that can be extracted could be the jpeg compressed versions of the uncompressed images (8bit) that can be viewed using the (expensive) manufacturer's software. So in some instances you are working with data that are compressed from linear to 8bit images (0-255) and then further compressed to jpeg when extracted from the DICOM file.
In general, you should try your best to work with the data relevant to the purpose you are pursuing. Don't worry too much about it, usually the whole research community experiences the same limitations that you are experiencing.
It depends on the type of processing and analysis you want to perform.
Please read this article:
Logarithmic image processing: The mathematical and physical framework for the representation and processing of transmitted images
I am agree with Larbi Messaouda ; It depends on the type of processing and analysis you want to perform.
It depends on the type of processing and analysis you want to perform.
Dear researchers
Logarithmic image processing: The mathematical and physical framework for the representation and processing of transmitted images
regards
Dear researchers
Logarithmic image processing: The mathematical and physical framework for the representation and processing of transmitted images
regards
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Human-Computer Interaction