Hello!
I have computer tomography images of human brain:
Be careful! Links below contains sensitive materials that can invoke vomiting and dizziness.
GIF image 1: Screen capture - ab24a491bb1af3437b73c410ba772324 - Gyazo
I would like to calculate contour of these things in the right:
image 1:
Can I get the same contours using image processing for every single image?
If yes, than that kind of algorithms should I look at?
If no, can I do it with neural networks? Are two gif’s I provided by me sufficient raw data for learning to get good result?
If no, how many of them do I need?
I am beginner in the OpenCV. I make this for university work.
I am sorry for so many questions 
. Will be glad for any help!
I cant embed more then two links, so additional links are here:
GIF image 2: Screen capture - 24b3488daabe6e3a052b4a3f2cd35f5f - Gyazo
image 2:
(don’t worry about tomographic images causing anyone around here distress.)
the circled area in the first image does not stand out to me. with domain knowledge, I might be able to read the image and point out the area/volume that displaces the functional brain matter. still, “traditional” approaches won’t solve this.
you’ll need deep learning, if you want a computer to mark these areas.
you would train a DNN to perform segmentation. you will need labeled training data. the data can be 2D (slices) or 3D (volumes), but it needs to be annotated/labeled. I would recommend volumetric data. if you choose to go with 2D data, the slices should be from the same plane in all samples, or they should contain some information on the plane’s height (assuming transverse plane).
1 Like
