-
Notifications
You must be signed in to change notification settings - Fork 3
Preprocessing
Now that we have a dataset in the right format (NIFTI), we still have some preprocessing steps to do before we can start.
An isotrope image is an uniform image in all orientations. The original voxel spacing of the images is (0.513393 x 0.513393 x 0.800003) and we want to have something looking like (A x A x A). I choose to take the minimum rounded value as A so my voxel spacing is (0.5 x 0.5 x 0.5).
You can use the script isotrope located in the tools folder which is using DIPY.
python3 tools/isotrope.py ../Datasets/Bullitt ../Datasets/Bullitt_Isotrope
The value of each voxel is an uint16 with a value between 0 and 1323.
log2(1323) = 10.36 so we need at least 11 bits per voxel.
The original image size is 448 x 448 x 128.
The isotrope image size is 460 x 460 x 205.
We apply feature standardization to the train, test and validation images. Feature standardization makes the values of each feature in the data have zero-mean (when subtracting the mean in the numerator) and unit-variance.
More information on feature scaling.