Augmentor3D

A module for 3D image augmentations for deep learning, specifically medical images such as CT, MRI.

Dependencies

SimpleITK

Please make sure the image headers (origin, spacing and direction) are correct, since the transformations completely depending on those headers.

3D Augmentation module examples

A module implemented for the purpose of augmenting 3D volumes.
This class uses transformations in SimpleITK library to perform augmentations.

from augmentation3DUtil import Augmentation3DUtil
from augmentation3DUtil import Transforms
import SimpleITK as sitk
import matplotlib.pyplot as plt
%matplotlib inline

Reading image using SimpleITK module

img = sitk.ReadImage(fr"**********filename***************")

Performing single transformations

Use the Enum class Transforms for defining transforms. See examples below

Rotation

parameters
probability : probability of the executing this particular transformation
degrees : degree of rotation in xy axis.

au = Augmentation3DUtil(img,mask=None)
au.add(Transforms.ROTATE2D,probability = 1, degrees = 45)

img,re = au.process(1)

arr = sitk.GetArrayFromImage(img)
augarr = sitk.GetArrayFromImage(re[0][0])

plt.subplot(121)
plt.imshow(arr[10])
plt.axis('off')
plt.subplot(122)
plt.imshow(augarr[10])
plt.axis('off')

(-0.5, 383.5, 383.5, -0.5)

Flip horizontal

parameters
probability : probability of the executing this particular transformation

au = Augmentation3DUtil(img,mask=None)
au.add(Transforms.FLIPHORIZONTAL,probability = 1)

img,re = au.process(1)

arr = sitk.GetArrayFromImage(img)
augarr = sitk.GetArrayFromImage(re[0][0])

plt.subplot(121)
plt.imshow(arr[10])
plt.axis('off')
plt.subplot(122)
plt.imshow(augarr[10])
plt.axis('off')

(-0.5, 383.5, 383.5, -0.5)

Flip vertical

parameters
probability : probability of the executing this particular transformation

au = Augmentation3DUtil(img,mask=None)
au.add(Transforms.FLIPVERTICAL,probability = 1)

img,re = au.process(1)

arr = sitk.GetArrayFromImage(img)
augarr = sitk.GetArrayFromImage(re[0][0])

plt.subplot(121)
plt.imshow(arr[10])
plt.axis('off')
plt.subplot(122)
plt.imshow(augarr[10])
plt.axis('off')

(-0.5, 383.5, 383.5, -0.5)

Translation

parameters
probability : probability of the executing this particular transformation
offset : 3D cordinates in mm ex. (5,5,0)

au = Augmentation3DUtil(img,mask=None)
au.add(Transforms.TRANSLATE,probability = 1, offset = (15,15,0))

img,re = au.process(1)

arr = sitk.GetArrayFromImage(img)
augarr = sitk.GetArrayFromImage(re[0][0])

plt.subplot(121)
plt.imshow(arr[10])
plt.axis('off')
plt.subplot(122)
plt.imshow(augarr[10])
plt.axis('off')

(-0.5, 383.5, 383.5, -0.5)

Shear

parameters
probability : probability of the executing this particular transformation
magnitude : 3D cordinates

au = Augmentation3DUtil(img,mask=None)
au.add(Transforms.SHEAR,probability = 1, magnitude = (0.1,0.1))

img,re = au.process(1)

arr = sitk.GetArrayFromImage(img)
augarr = sitk.GetArrayFromImage(re[0][0])

plt.subplot(121)
plt.imshow(arr[10])
plt.axis('off')
plt.subplot(122)
plt.imshow(augarr[10])
plt.axis('off')

(-0.5, 383.5, 383.5, -0.5)

Random sampling of transformations (combination)

specify the number of samples to be obtained using "process" method
specify the likelihood of a particular transform as probabilities.

au = Augmentation3DUtil(img,mask=None)
au.add(Transforms.SHEAR,probability = 0.25, magnitude = (0.05,0.05))
au.add(Transforms.TRANSLATE,probability = 0.75, offset = (15,15,0))
au.add(Transforms.ROTATE2D,probability = 0.75, degrees = 10)
au.add(Transforms.FLIPHORIZONTAL,probability = 0.75)

img,re = au.process(5)

for i in range(len(re)):
    arr = sitk.GetArrayFromImage(img)
    augarr = sitk.GetArrayFromImage(re[i][0])

    plt.subplot(121)
    plt.imshow(arr[10])
    plt.axis('off')
    plt.subplot(122)
    plt.imshow(augarr[10])
    plt.axis('off')
    plt.show()