I’ve been taking the fastai course and reading several threads of this forum in order to find possible approaches to a current problem I’m focusing on.
I’d like to detect the probability of breast cancer from a whole and high resolution image. As most of all the traditional models can handle low resolution images it’s difficult for me to find a way to approach this.
Suppose that all I have is a big dataset consisting on: a high res microscopy image and a label (cancer/not cancer for the sake of simplicity).
What would you do to greatly decrease the image resolution without losing too much detail, probably hurting the accuracy?
I’ve read that an approach suugest by Jeremy would be to:
- Build a 1st model to find “interesting areas” on each image.
- Crop out high res versions of small areas around each of those interesting areas and make that input to a 2nd model.
But I have no idea how to implement this. Any examples about something similar to this?