Thanks for the great work. I'm confused about how is the fine-grained tumor classification (Figure 3) mentioned in the paper performed? Since you cropped images to 96x96x96 from pre-training, I imagine you are using ROI with same size that contain tumors to perform this task. Is it correct?