niftynet.network.dense_vnet module¶

DenseVNetDesc¶: alias of DenseVNetParts

class DenseVNet(num_classes, hyperparameters={}, architecture_parameters={}, w_initializer=None, w_regularizer=None, b_initializer=None, b_regularizer=None, acti_func='relu', name='DenseVNet')[source]¶

Bases: niftynet.network.base_net.BaseNet

implementation of Dense-V-Net:: Gibson et al. Automatic multi-organ segmentation on abdominal CT with dense V-networks

### Diagram

DFS = Dense Feature Stack Block

Initial image is first downsampled to a given size.
Each DFS+SD outputs a skip link + a downsampled output.
All outputs are upscaled to the initial downsampled size.
If initial prior is given add it to the output prediction.

Input

–[ DFS ]———————–[ Conv ]————[ Conv ]——[+]–>

| | |

—–[ DFS ]—————[ Conv ]—— | |

| |

—–[ DFS ]——-[ Conv ]——— |: [ Prior ]—

The layer DenseFeatureStackBlockWithSkipAndDownsample layer implements [DFS + Conv + Downsampling] in a single module, and outputs 2 elements:

Skip layer: [ DFS + Conv]

Downsampled output: [ DFS + Down]

create_network()[source]¶

downsample_input(input_tensor, n_spatial_dims)[source]¶

layer_op(input_tensor, is_training, layer_id=-1)[source]¶

image_resize(image, output_size)[source]¶

class SpatialPriorBlock(prior_shape, output_shape, name='spatial_prior_block')[source]¶

Bases: niftynet.layer.base_layer.TrainableLayer

layer_op()[source]¶

DenseFSBlockDesc¶: alias of DenseFSDesc

class DenseFeatureStackBlock(n_dense_channels, kernel_size, dilation_rates, use_bdo, name='dense_feature_stack_block', **kwargs)[source]¶

Bases: niftynet.layer.base_layer.TrainableLayer

Dense Feature Stack Block

Stack is initialized with the input from above layers.
Iteratively the output of convolution layers is added to the stack.
Each sequential convolution is performed over all the previous stacked channels.

Diagram example:

stack = [Input] stack = [stack, conv(stack)] stack = [stack, conv(stack)] stack = [stack, conv(stack)] … Output = [stack, conv(stack)]

create_block()[source]¶

layer_op(input_tensor, is_training=None, keep_prob=None)[source]¶

DenseSDBlockDesc¶: alias of DenseSDBlock

class DenseFeatureStackBlockWithSkipAndDownsample(n_dense_channels, kernel_size, dilation_rates, n_seg_channels, n_downsample_channels, use_bdo, name='dense_feature_stack_block', **kwargs)[source]¶

Bases: niftynet.layer.base_layer.TrainableLayer

Dense Feature Stack with Skip Layer and Downsampling

Downsampling is done through strided convolution.

—[ DenseFeatureStackBlock ]———-[ Conv ]——- Skip layer: ——————– Downsampled Output

See DenseFeatureStackBlock for more info.

create_block()[source]¶

layer_op(input_tensor, is_training=None, keep_prob=None)[source]¶

class AffineAugmentationLayer(scale, interpolation, boundary, transform_func=None, name='AffineAugmentation')[source]¶

Bases: niftynet.layer.base_layer.TrainableLayer

This layer applies a small random (per-iteration) affine transformation to an image. The distribution of transformations generally results in scaling the image up, with minimal sampling outside the original image.

__init__(scale, interpolation, boundary, transform_func=None, name='AffineAugmentation')[source]¶

” scale denotes how extreme the perturbation is, with 1. meaning

no perturbation and 0.5 giving larger perturbations.

interpolation denotes the image value interpolation used by: the resampling

boundary denotes the boundary handling used by the resampling transform_func should be a function returning a relative transformation (mapping <-1..1,-1..1,-1..1> to <-1..1,-1..1,-1..1> or <-1..1,-1..1> to <-1..1,-1..1>)