0

I have a 3D CT image of a car engine in raw form with no header information. When loaded into numpy array as 16 bit unsigned integer, I noticed that the values range between 0 to 52000. Is this normal for a CT Image? When viewing the image, I also noticed a lot of cloud like noise in every slice. I'm trying to extract features using a deep learning method. This is my first time working with CT Images. What pre processing is neccesary for such CT Images?

faiz fz
  • 41
  • 6
  • Welcome to StackOverflow, Please take this [Tour](https://stackoverflow.com/tour), and read [How to Ask](https://stackoverflow.com/help/how-to-ask). Please [Edit](https://stackoverflow.com/posts/66229988/edit) this post to include your own effort into solving the problem, AKA the [Minimal Reproducible Example](https://stackoverflow.com/help/minimal-reproducible-example) – Bilal Feb 16 '21 at 18:47

2 Answers2

0

Since it is 16 uint format, the value range could be 0 ~ 2 ^ 16 which is up to 65535, so 0 to 52000 is very reasonable. If it is a car engine instead of soft tissue, chances are you don't need that big range, you could reduce your data to 12 bit or even 8 bit without losing much details, by applying windowing and leveling to it.

For the "cloud like noise", please upload some image to show exact problem. If my guess is correct, it is probably due to the endian-ness. Try opening the raw data using ImageJ and try 16 uint little-endian and big-endian, one of these might give you normal image.

Guang
  • 355
  • 1
  • 8
0

From my experience in working with soft tissue CT medical images, I have seen CT having -1024 intensity for air and up to +3000 intensity for cancellous bone. When it comes to metal-like electrodes and even tooth images, it reaches up to +9000. So I guess for heavy metals like car engines the intensity value of 52000 is not aberration(though I never heard CT images of car engines so far!).

Pre-processing can include windowing or normalization operations and noise removals. Windowing is something already mentioned here to restructure the intensity range to a more workable one.

banikr
  • 63
  • 1
  • 9