Segmentation in Medical Images
Hello everyone, I am pretty new to machine learning. After couple days of searching, I found fast ai, and I watched couple courses on youtube to learn more about machine learning. The course talked heavily about classification and segmentation. But I want to build something that can find difference between two similar images (similar content but can’t just use openCV to find difference, so maybe the brightness of the two are slightly different, there might be shift between the two, and color might be slightly different, etc). So I began to look into using machine learning, after all, machine learning is doing really well in image field.
What I want to ask is, what kind of framework should I use? How do I give two images? (most of the CNN network takes one?) I looked into U-net and found couple posts here about segmentation for medical use, such as cancer finding. I found a competition called DAGM 2007, and the challenge is about finding the defect (industrial optical inspection). I don’t know if this is good enough for my project, because for my project, I just need to find the difference between the test image and golden image, and I would imaging being able to use golden image can provide more information.
How should I approach this problem? segmentation using U-net? I also found something called attentive recurrent comparators, which uses RNN to find the difference. (which I imaging will serve better if I have a golden image?)