Crop

crop[source]

crop(image_array, final_dims_in_pixels, zero_fill_mode=False)

Crop image data to final_dim_in_pixels

Attributes: image_array (float, np.array): 3D numpy array containing image data final_dim_in_pixels (list): Final number of pixels in xyz dimensions. Example: [256, 256, 80] zero_fill_mode (bool): If True, returns array filled with zeros

Returns: cropped_image_array (arr): Resized array containing image data

Usage

crop takes an image array and crops the array according to the input dimensions.

See the example below:

import pkg_resources

DATA_ROOT = pkg_resources.resource_filename('I2T2', 'data')

# test
knee_data_path = os.path.join(DATA_ROOT,'knee')

dcm_df = dicom_dataframe(path_to_dicom_dir=knee_data_path, dicom_extension='dcm')
image_array = dcm_df.get_pixel_data()
cropped_image = crop(image_array=image_array,
                     final_dims_in_pixels=[400, 400, 22])
cropped_zero_image = crop(image_array=image_array, final_dims_in_pixels=[
                          400, 400, 22], zero_fill_mode=True)

# original image
plt.figure(figsize=(16, 16))
plt.subplot(131)
plt.title('Original image')
plt.imshow(image_array[:, :, 6])

# cropped image
plt.subplot(132)
plt.title('Cropped image')
plt.imshow(cropped_image[:, :, 6])

# cropped image zero'd
plt.subplot(133)
plt.title("Cropped zero'd image")
plt.imshow(cropped_zero_image[:, :, 6])

plt.show()

Pad

pad[source]

pad(image_array, final_dims_in_pixels, zero_fill_mode=False)

Pad image data to final_dim_in_pixels

Attributes: image_array (float, np.array): 3D numpy array containing image data final_dim_in_pixels (list): Final number of pixels in xyz dimensions. Example: [256, 256, 80] zero_fill_mode (bool): If True, returns array filled with zeros

Returns: padded_image_array (arr): Resized array containing image data

Usage

pad takes an image array and crops the array according to the input dimensions.

See the example below:

dcm_df = dicom_dataframe(path_to_dicom_dir=knee_data_path, dicom_extension='dcm')

image_array = dcm_df.get_pixel_data()

padded_image = pad(image_array=image_array,
                   final_dims_in_pixels=[800, 800, 22])
padded_zero_image = pad(image_array=image_array, final_dims_in_pixels=[
                        800, 800, 22], zero_fill_mode=True)

# original image
plt.figure(figsize=(16, 16))
plt.subplot(131)
plt.title('Original image')
plt.imshow(image_array[:, :, 6])

# padded image
plt.subplot(132)
plt.title('Padded image')
plt.imshow(padded_image[:, :, 6])

# padded image zero'd
plt.subplot(133)
plt.title("Padded zero'd image")
plt.imshow(padded_zero_image[:, :, 6])

plt.show()

Resample

resample[source]

resample(image_array, target_pixel_dims_list, is_seg=False)

Resamples image to new resolution according to a compression factor using scipy's ndimage

Attributes: image_array (float, np.array) : Array of voxel values for image target_pixel_dims_list (list) : Dimensions of final image in pixels (e.g. [256, 256, 128]) is_seg (bool) : Whether or not the image is a segmentation (i.e. pixel values are int)

Returns: resampled_image_array (arr): Resampled array containing image data

resample_by[source]

resample_by(image_array, compression_factor_list, is_seg=False)

Resamples image to new resolution according to a compression factor using scipy's ndimage

Attributes: image_array (float, np.array) : Array of voxel values for image compression_factor_list (list) : Compression or expansion factor (e.g. [0.1, 10.0, 1.0]) is_seg (bool) : Whether or not the image is a segmentation (i.e. pixel values are int)

Returns: resampled_image_array (arr): Resampled array containing image data

Usage

resample takes an image array and resamples it to the dimensions given in target_pixel_dims_list.

resample_by is similar to resample but user gives a compression factor instead (e.g. [0.1,0.1,1.0] will resample the image to 10% in x and y directions)

See the examples below:

dcm_df = dicom_dataframe(path_to_dicom_dir=knee_data_path, dicom_extension='dcm')

image_array = dcm_df.get_pixel_data()

downsampled_image = resample(image_array=image_array, target_pixel_dims_list=[
                             51, 51, 22], is_seg=False)

upsampled_image = resample(image_array=downsampled_image, target_pixel_dims_list=[
                           512, 512, 22], is_seg=False)

# Plot
plt.figure(figsize=(16, 16))

plt.subplot(131)
plt.title('Original image')
plt.imshow(image_array[:, :, 6])

plt.subplot(132)
plt.title('Downsampled image')
plt.imshow(downsampled_image[:, :, 6])

plt.subplot(133)
plt.title('Re-upsampled image')
plt.imshow(upsampled_image[:, :, 6])

plt.show()

Example of using resample_by:

dcm_df = dicom_dataframe(path_to_dicom_dir=knee_data_path, dicom_extension='dcm')

image_array = dcm_df.get_pixel_data()

downsampled_image = resample_by(image_array=image_array,
                                compression_factor_list=[0.1, 0.1, 1.0],
                                is_seg=False)

upsampled_image = resample_by(image_array=downsampled_image,
                              compression_factor_list=[10, 10, 1.0],
                              is_seg=False)

# Plot
plt.figure(figsize=(16, 16))

plt.subplot(131)
plt.title('Original image')
plt.imshow(image_array[:, :, 6])

plt.subplot(132)
plt.title('Downsampled image')
plt.imshow(downsampled_image[:, :, 6])

plt.subplot(133)
plt.title('Re-upsampled image')
plt.imshow(upsampled_image[:, :, 6])

plt.show()

Resampling example:

Resampling segmentation to agree with original image

# read data
segmentation_path = os.path.join(DATA_ROOT,'knee-segmentation','knee_segmentation.h5')

dcm_df = dicom_dataframe(path_to_dicom_dir=knee_data_path, dicom_extension='dcm')

image_array = dcm_df.get_pixel_data()
knee_segmentation = load_h5(segmentation_path)
knee_segmentation = knee_segmentation.get('seg')

print('Knee segmentation dims:', knee_segmentation.shape)
print('Knee original dims:', image_array.shape)

Knee segmentation dims: (280, 280, 160)
Knee original dims: (512, 512, 22)

# resample to desired shape
resampled_segmentation = resample(image_array=knee_segmentation,
                                  target_pixel_dims_list=(512, 512, 22),
                                  is_seg=True)

print('Knee segmentation dims after resampling:', resampled_segmentation.shape)

Knee segmentation dims after resampling: (512, 512, 22)

# Plot
from matplotlib.colors import ListedColormap
plt.figure(figsize=(16, 16))

plt.subplot(131)
plt.title('Original image')
plt.imshow(image_array[:, :, 6])

plt.subplot(132)
plt.title('Segmentation image')
plt.imshow(resampled_segmentation[:, :, 6])

colors = ['white', 'purple', 'red', 'green', 'blue']
cmap = ListedColormap(colors)

plt.subplot(133)
plt.title('Overlaid segmentation')
# I would add interpolation='none'
plt.imshow(image_array[:, :, 6], cmap='gray')
plt.imshow(resampled_segmentation[:, :, 6], cmap=cmap,
           vmin=0, vmax=4, alpha=0.6)  # interpolation='none'

plt.show()

Resize

resize[source]

resize(image_array, original_pixel_spacing_list, target_field_of_view_list=[140, 140, 80], target_pixel_spacing_list=[0.5, 0.5, 0.5], is_seg=False)

Resizes image_array to desired field of view by first resampling and then cropping / padding. The returned resampled image has isotropic pixel spacing = 0.5mm

Args: image_array: np.array of voxel values pixel_spacing_list: list of pixel spacing values prior to resizing in mm target_field_of_view_list: list of field of values in mm (default: 140, 140, 80) target_pixel_spacing_list: list of pixel spacing values after resizing in mm (default: 0.5, 0.5, 0.5) is_seg (bool): whether image_array is a segmentation or not (default: False)

Returns: image_cropped_resampled: np.array with voxels values after cropping and re-sampling

Resizing image

# read data
dcm_df = dicom_dataframe(path_to_dicom_dir=knee_data_path, dicom_extension='dcm')

image_array = dcm_df.get_pixel_data()
knee = load_h5(segmentation_path)
knee_image = knee.get('img')
knee_segmentation = knee.get('seg')

print('Knee segmentation dims:', knee_segmentation.shape)
print('Knee resampled dims:', knee_image.shape)
print('Knee original dims:', image_array.shape)

Knee segmentation dims: (280, 280, 160)
Knee resampled dims: (280, 280, 160)
Knee original dims: (512, 512, 22)

xres, yres = _get_tag_from_loaded_dicom(dcm_df.dataframe['DS'].loc[0], 'PixelSpacing')
zres = _get_tag_from_loaded_dicom(dcm_df.dataframe['DS'].loc[0], 'SpacingBetweenSlices')
xres, yres, zres = float(xres), float(yres), float(zres)
resized_image = resize(image_array=image_array, original_pixel_spacing_list=[xres, yres, zres], target_pixel_spacing_list=[0.5, 0.5, 0.5])

Cropping image in xy
Cropping image in z

# Plot
from matplotlib.colors import ListedColormap
plt.figure(figsize=(16, 16))

plt.subplot(131)
plt.title('Original image')
plt.imshow(image_array[:,:,10])

plt.subplot(132)
plt.title('Resized image')
plt.imshow(resized_image[:,:,72])

colors = ['white', 'purple', 'red', 'green', 'blue']
cmap = ListedColormap(colors)

plt.show()