1

I have a dataset which contains 3D CT scans from different patients along with the segmenation masks of a certain organ. The 3D scans have been drawn each day for a period of 30 days for each patient. I want to conduct a series of experiments and measure the similarity of the organ for each patient over this period. There is no need to tackle this dataset as a whole, all I need is to find the pairwise correlation of the organ from the first day and on.

Is there any technique to find the correlation between two 3D objects?

Dimimal13
  • 85
  • 8
  • What is the metric you are using to measure the similarity of the organ? If you don't have one and just want to compare the images then I don't think you are going to be able to use a metric so simple as correlation coefficient. You might be able to use the cosine similarity between the matrices of numeric pixel values between images *if* your images are very clean and consistent. – bstrain Jul 01 '20 at 16:58
  • I don't think there is a similarity metric for an organ. And I actually use the segmentation masks to obtain the 3D structure of the organ, so its a binary 3D voxel. So the best approach is to deal with them as matrices and measure the cosine similarity? – Dimimal13 Jul 01 '20 at 18:17
  • I am not sure if it is the best approach, but it is one approach where you can measure the differences in the absence of another metric. – bstrain Jul 01 '20 at 21:52
  • Can you add an example of a few rows of your dataset? – fractalnature Jul 10 '20 at 17:09
  • @fractalnature What do you mean by a few rows of the dataset? Do you mean to put a 3D sample here? – Dimimal13 Jul 11 '20 at 13:22
  • You haven't transformed the pixels from the scans into a data frame? – fractalnature Jul 11 '20 at 19:39
  • @fractalnature Actually I have them stored in a numpy file. Even if I plot them here, I don't think if you can make any conclusions – Dimimal13 Jul 11 '20 at 20:13
  • @Dimimal13 I didn't mean plot the data or put it here so that we can make conclusions from it. Its more just a way for us to understand what you are working with. Its helpful to know how your data is formatted. What are the columns? what does the data look like? All of these things are helpful when trying to provide advice. see this question: https://datascience.stackexchange.com/questions/47384/artificially-increasing-frequency-weight-of-word-ending-characters-in-word-build – fractalnature Jul 11 '20 at 22:03
  • @fractalnature I see, the data are actually binary masks in the 3D space. The problem is that each of them has a different size in the Z axis so even if an object is similar with the other one, you have to match it in the Z space as well.... That makes the problem even harder. – Dimimal13 Jul 12 '20 at 17:47

0 Answers0