Hi all,
I'm hoping you could help me with a conundrum regarding reduction of size of dicom CT pixeldata.
I'm using the Cornerstone DICOM parser (https://github.com/cornerstonejs/dicomParser) to extract pixeldata from CT DICOM images which is working fine.
I need however, to reduce the dimensions of the obtained pixeldata; i.e. from 512*512 to 256*256 while maintaining as much image quality as possible. So I need to do some kind of interpolation algorithm that works on CT pixeldata (i.e. range from 0-approx4096) instead of just skipping pixels. After image size reduction CT window/level needs to be applied. Could anyone point me in the direction of an algorithm that works in regard to reduce the dimensions of pixeldata with HU values?
Sincerely
I think averaging the four pixels that become one pixel, is an easy good option. You will then reduce the noise so it is better than skipping pixels. Don't forget to adjust the coordinates.
You might consider 'locally adaptive' averaging: average only when there is no high gradient present. If there is a high gradient, take the median pixelvalue. Such adaptive averaging is more edge preserving, at the cost of somewhat more noise.
Depends on how much computation you are willing to perform. Since you are using a HTML5 based parser you may be limited in the amount of computation available locally. If you are looking for high quality image scaling, I like b-splines for multi-axis fits in image space. I don't know a Nodejs library but I can point you to a Java library that may be useful. Erik Meijering wrote a nice java library several years ago. See the link below:
https://imagescience.org/meijering/software/transformj/
It is java and written for another appliation but will give you a chance to test the algoritm before investing any effort. You can deduce the algorithm from the source or use a subset of the java code as a library in your application.
Mr Morris and Mr Zipj,
Thank you for taking the time to answer my enquiry.
I went with bilinear interpolation.
Please see the following URLS with proof of concepts:
http://castlemountain.dk/dicomParser/indexBilin.html
http://castlemountain.dk/dicomParser/indexBilin2.html
Here I have used the cornerstone DICOM parser to extract pixeldata from a 512x512 CT non-contrast abdominal axial slice.
Take a look at "image 2" in the 2 URLS.
Image 2 has been reduced in size to 256*256 (compression factor 2).
In the first URL the pixeldata has been subjected to window/level (javascript function 'convertToGreylevel') prior to using a bilinear interpolation reduction algorithm (javascript function 'bilin') which generates a new array of pixeldata which is subsequently used to draw the 256x256 slice (i.e. image 2).
In the second URL the extracted pixeldata is reduced in size with the 'bilin' function referenced above and after the 256*256 reduced pixelarray has been obtained window/level is applied to the pixeldata.
It is apparent that image 2 when window/level is applied after bilinear interpolation is more coarse and I can't seems to figure out why that is. To@ the best of my reasoning it shouldn't make a difference if bilinear interpolation is applied prior to window/level or vice versa!?
Could you guys perhaps lend some more of your time to enlighten me?
It should not make a difference.
I guess bilinear interpolation amounts to the same thing as averaging. Averaging would be faster though.
Mr Zijp,
Yes averaging is faster but I will also need to do scaling with different compression factors.
You agree that it shouldn't make a difference if window/level is prior or after bilinear interpolation....so looking at the above URLs which implements bilinear interpolation prior and after w/l, repectively, can you point to the reason why the latter is more coarse?
- Badges
- Science topic
- Similar topics
- Computer Science
- Images