Currently, some input images are rotated by 180 degrees. Should do:
[ ] Standardize angle of orientation for each input image (probably can get relevant info from the dcm file)
[ ] Standardize the zoom for each input image: create a bounding box around the relevant lung area.
Kyle's lung segmentation model will likely be helpful for this.
[x] Add more to-do boxes to check off
After standardizing orientation, may be interesting to try on-the-fly augmentation of images, e.g. doubling the training set size by creating mirror images of each instance.
Currently, some input images are rotated by 180 degrees. Should do:
dcmfile)After standardizing orientation, may be interesting to try on-the-fly augmentation of images, e.g. doubling the training set size by creating mirror images of each instance.