In this question, I would like to focus on intensity value of the CT Scan. First, take a look at the image below:

lung nodule measurement

The upper image is the original image, while the lower image is the thresholded version. In order to measure volume of any shape, theoretically, it is possible to simply just count the number of voxels in the image. However, the outermost layer of the object (ex. nodule) shows darker intensity, while all voxels inside the object have very high intensity. If I simply count the voxels in the thresholded version, I will highly likely to get a result volume larger than the actual volume for the lung nodule.

I also see that there are variables such as window center (level) and window width, which can be used to adjust the DICOM image's intensity information. Different intensity can alter the result volume.

So here is the question: if I am to measure any given lung nodule, what should I do in order to achieve the best possible precision? When should we ignore the lower intensity voxels? Or must I do this in some other ways?

  • $\begingroup$ Hope you don't mind, but I moved this over to Signal Processing, because it's more of a platform-independent processing issue, and I've seen questions like this get great answers over here. $\endgroup$ Aug 8, 2012 at 18:43
  • $\begingroup$ Is there a reason you're thresholding it first? You'd probably get more accurate results without that. $\endgroup$
    – endolith
    Aug 27, 2012 at 21:57
  • $\begingroup$ Well, then which voxel should you take into account? $\endgroup$
    – Karl
    Aug 28, 2012 at 3:46

2 Answers 2


Aside from the whole discussion of pure signal processing: What exactly do you define as "the nodule". This is usually a biological entity with difficult to define borders. The nature of the nodule is sometimes invasive growth and therefore even in histological sections poorly defined. The CT itself has therefore a threshold higher than histology and therefore the true border of a nodule cannot defined easily. On the other hand, healthy tissue around the nodule might be compressed and appear dense in CT. This might be depending on the respiration phase the scan was triggert on (best results in midinspiration, according to newer data). Or inflammation might obscure the true border of the nodule.

Another aspect is interpolation of the images. The technique is usually a spiral-CT, so you don´t miss any lesions. The sections are therefore calculated. This results less defined borders of objects. If the focus would be on border detection or near histology resolution, high-resolution scans are usually required. This is what you do for interstitial lung disease. Unfortunately, they slice through the lung with a pretty large distance in between. Radiation dose would be extremely high, if you try to get a "full volume scan". But in this type of scan, you have to be aware, that You´s miss lesions between slices.

To come back to the initial question: I think you will have to validate your technique - whatever it is - with the gold standard. Which is histology sections. (Unfortunately, the lung isn´t easily sectioned...). Another option is an additional technique: like PET-CT (combining positron emission tomography with computer tomography), but the alignment algorithms are sometimes tricky.


Assuming this is due to the "partial volume effect" (and not because the outer layer of the nodule really is some different material):

If you consider a voxel (1mm^3, say) with a bright value of say 200 to be all nodule material, and a voxel with a value of say 100 to be definitely normal tissue, then it seems reasonable to assume that a voxel with a value of 160 is 60% nodule and 40% normal tissue (and so it should contribute 0.6mm^3 to your total).

If that assumption is correct (and that's a big if), then it should get you a better measurement of volume than just counting voxels >=200 or >100.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.