Giters

DIDSR / VICTRE

Virtual Imaging Clinical Trial for Regulatory Evaluation

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Post processing of DM images

AngieNicole-Hernandez opened this issue · comments

commented

Hello, I have a question about the visualization of the MD images. I am interested in obtaining a sort of "post processed" DM image, such as the one presented in one of the VICTRE articles:

image

I saw, in another issue, that the lesions were made more conspicuous by increasing their radiography attenuation during imaging, I think that might mean a modification of the "density" of the material indexed by 200 in the SECTION MATERIAL FILE LIST on the .in files for imaging, is this correct? Can I get more details about increasing their radiography attenuation during imaging ?

The below image is one provided, in the repository for X-ray imaging, that has a lesion (mcgpu_image_22183101_scattered_0000.raw), I played with the brightness and window level, however the inner structures of the breast or the mass are not easily identifiable.

image

Any extra information I can get about the post processing of the images for display purposes would be of great help. Thank you very much!

commented

Dear Nicole,

We have not developed software to improve the visualization of the entire mammogram with a fixed window/level, as it is done in clinical practice to transform the raw data in "for processing" mode to "for presentation" mode. This is a limitation of our work, but we think it is justified because we used only model observers to detect the lesions. It is generally assumed (but not warrantied) that the image post-processing that is helpful for radiologists reading the images on a display might not be necessary for algorithms (we don't model the display effects neither).

Some of the images in our papers were post-processed with a simple log scale and histogram equalization to improve the visibility in the printed article, and this should be mentioned in the figure caption.

The lesions are very hard to see in most phantoms because they have almost the same attenuation as glandular tissue (same composition and about 5% higher density only, as reported in the bibliography). To artificially increase the density of the lesion, the method you mention is correct: increase the material density in the input file (https://github.com/DIDSR/VICTRE_MCGPU/blob/5ffd1db0e21c01137bba0dacd877986b58fa2a13/MC-GPU_v1.5b_sample_mammo_and_DBT_simulation_InputDensity.in#L73).
You could also model a completely calcified lesion assigning to the lesion the same material and density as a calcification, but this will give you a very unrealistic contrast.

These are a couple of interesting papers for the conversion of mammograms to "for presentation":

I hope this helps.

   Andreu