import tensorflow
from PrognosAIs.Model.Architectures.Architecture import ClassificationNetworkArchitecture
from tensorflow.keras.layers import Add
from tensorflow.keras.layers import BatchNormalization
from tensorflow.keras.layers import Conv2D
from tensorflow.keras.layers import Conv3D
from tensorflow.keras.layers import Dense
from tensorflow.keras.layers import Dropout
from tensorflow.keras.layers import GlobalAveragePooling2D
from tensorflow.keras.layers import GlobalAveragePooling3D
from tensorflow.keras.layers import Input
from tensorflow.keras.layers import MaxPooling2D
from tensorflow.keras.layers import MaxPooling3D
from tensorflow.keras.layers import ReLU
from tensorflow.keras.models import Model
[docs]class ResNet(ClassificationNetworkArchitecture):
[docs] def get_residual_conv_block(
self, layer: tensorflow.keras.layers, N_filters: int, kernel_size: list
):
N_dims = len(kernel_size)
if N_dims == 2:
conv_func = Conv2D
elif N_dims == 3:
conv_func = Conv3D
stride_size = self.get_corrected_stride_size(layer, [2] * N_dims, kernel_size)
conv_1 = conv_func(
filters=N_filters, kernel_size=kernel_size, strides=stride_size, padding="same",
)(layer)
batchnorm_1 = BatchNormalization(axis=-1)(conv_1)
relu_1 = ReLU()(batchnorm_1)
conv_2 = conv_func(
filters=N_filters, kernel_size=kernel_size, strides=[1] * N_dims, padding="same"
)(relu_1)
batchnorm_2 = BatchNormalization(axis=-1)(conv_2)
conv_skip = conv_func(
filters=N_filters, kernel_size=[1] * N_dims, strides=stride_size, padding="valid"
)(layer)
batch_skip = BatchNormalization(axis=-1)(conv_skip)
res_output = Add()([batch_skip, batchnorm_2])
relu_output = ReLU()(res_output)
return relu_output
[docs] def get_residual_identity_block(self, layer, N_filters, kernel_size):
N_dims = len(kernel_size)
if N_dims == 2:
conv_func = Conv2D
elif N_dims == 3:
conv_func = Conv3D
conv_1 = conv_func(
filters=N_filters, kernel_size=kernel_size, strides=[1] * N_dims, padding="same"
)(layer)
batchnorm_1 = BatchNormalization(axis=-1)(conv_1)
relu_1 = ReLU()(batchnorm_1)
conv_2 = conv_func(
filters=N_filters, kernel_size=kernel_size, strides=[1] * N_dims, padding="same"
)(relu_1)
batchnorm_2 = BatchNormalization(axis=-1)(conv_2)
res_output = Add()([batchnorm_2, layer])
relu_output = ReLU()(res_output)
return relu_output
[docs]class ResNet_18_2D(ResNet):
[docs] def create_model(self):
self.check_minimum_input_size(self.inputs, [9, 9])
stride_size = self.get_corrected_stride_size(self.inputs, [2, 2], [7, 7])
conv_1 = Conv2D(
filters=64, kernel_size=[7, 7], strides=stride_size, padding="valid", activation="relu",
)(self.inputs)
stride_size = self.get_corrected_stride_size(conv_1, [2, 2], [3, 3])
pooling_1 = MaxPooling2D(pool_size=[3, 3], strides=stride_size, padding="valid",)(conv_1)
res_block_1 = self.get_residual_identity_block(pooling_1, 64, [3, 3])
res_block_2 = self.get_residual_identity_block(res_block_1, 64, [3, 3])
res_block_2 = self.make_dropout_layer(res_block_2)
res_block_3 = self.get_residual_conv_block(res_block_2, 128, [3, 3])
res_block_4 = self.get_residual_identity_block(res_block_3, 128, [3, 3])
res_block_4 = self.make_dropout_layer(res_block_4)
res_block_5 = self.get_residual_conv_block(res_block_4, 256, [3, 3])
res_block_6 = self.get_residual_identity_block(res_block_5, 256, [3, 3])
res_block_6 = self.make_dropout_layer(res_block_6)
res_block_7 = self.get_residual_conv_block(res_block_6, 512, [3, 3])
res_block_8 = self.get_residual_identity_block(res_block_7, 512, [3, 3])
gap_1 = GlobalAveragePooling2D()(res_block_8)
gap_1 = self.make_dropout_layer(gap_1)
predictions = self.outputs(gap_1)
model = Model(inputs=self.inputs, outputs=predictions)
return model
[docs]class ResNet_34_2D(ResNet):
[docs] def create_model(self):
self.check_minimum_input_size(self.inputs, [9, 9])
stride_size = self.get_corrected_stride_size(self.inputs, [2, 2], [7, 7])
conv_1 = Conv2D(
filters=64, kernel_size=[7, 7], strides=stride_size, padding="valid", activation="relu",
)(self.inputs)
stride_size = self.get_corrected_stride_size(conv_1, [2, 2], [3, 3])
pooling_1 = MaxPooling2D(pool_size=[3, 3], strides=stride_size, padding="valid",)(conv_1)
res_block_1 = self.get_residual_identity_block(pooling_1, 64, [3, 3])
res_block_2 = self.get_residual_identity_block(res_block_1, 64, [3, 3])
res_block_3 = self.get_residual_identity_block(res_block_2, 64, [3, 3])
res_block_3 = self.make_dropout_layer(res_block_3)
res_block_4 = self.get_residual_conv_block(res_block_3, 128, [3, 3])
res_block_5 = self.get_residual_identity_block(res_block_4, 128, [3, 3])
res_block_6 = self.get_residual_identity_block(res_block_5, 128, [3, 3])
res_block_7 = self.get_residual_identity_block(res_block_6, 128, [3, 3])
res_block_7 = self.make_dropout_layer(res_block_7)
res_block_8 = self.get_residual_conv_block(res_block_7, 256, [3, 3])
res_block_9 = self.get_residual_identity_block(res_block_8, 256, [3, 3])
res_block_10 = self.get_residual_identity_block(res_block_9, 256, [3, 3])
res_block_11 = self.get_residual_identity_block(res_block_10, 256, [3, 3])
res_block_12 = self.get_residual_identity_block(res_block_11, 256, [3, 3])
res_block_13 = self.get_residual_identity_block(res_block_12, 256, [3, 3])
res_block_13 = self.make_dropout_layer(res_block_13)
res_block_14 = self.get_residual_conv_block(res_block_13, 512, [3, 3])
res_block_15 = self.get_residual_identity_block(res_block_14, 512, [3, 3])
res_block_16 = self.get_residual_identity_block(res_block_15, 512, [3, 3])
gap_1 = GlobalAveragePooling2D()(res_block_16)
gap_1 = self.make_dropout_layer(gap_1)
predictions = self.outputs(gap_1)
model = Model(inputs=self.inputs, outputs=predictions)
return model
[docs]class ResNet_18_3D(ResNet):
[docs] def create_model(self):
self.check_minimum_input_size(self.inputs, [9, 9, 9])
stride_size = self.get_corrected_stride_size(self.inputs, [2, 2, 2], [7, 7, 7])
conv_1 = Conv3D(
filters=64,
kernel_size=[7, 7, 7],
strides=stride_size,
padding="valid",
activation="relu",
)(self.inputs)
stride_size = self.get_corrected_stride_size(conv_1, [2, 2, 2], [3, 3, 3])
pooling_1 = MaxPooling3D(pool_size=[3, 3, 3], strides=stride_size, padding="valid",)(conv_1)
res_block_1 = self.get_residual_identity_block(pooling_1, 64, [3, 3, 3])
res_block_2 = self.get_residual_identity_block(res_block_1, 64, [3, 3, 3])
res_block_2 = self.make_dropout_layer(res_block_2)
res_block_3 = self.get_residual_conv_block(res_block_2, 128, [3, 3, 3])
res_block_4 = self.get_residual_identity_block(res_block_3, 128, [3, 3, 3])
res_block_4 = self.make_dropout_layer(res_block_4)
res_block_5 = self.get_residual_conv_block(res_block_4, 256, [3, 3, 3])
res_block_6 = self.get_residual_identity_block(res_block_5, 256, [3, 3, 3])
res_block_6 = self.make_dropout_layer(res_block_6)
res_block_7 = self.get_residual_conv_block(res_block_6, 512, [3, 3, 3])
res_block_8 = self.get_residual_identity_block(res_block_7, 512, [3, 3, 3])
gap_1 = GlobalAveragePooling3D()(res_block_8)
gap_1 = self.make_dropout_layer(gap_1)
predictions = self.outputs(gap_1)
model = Model(inputs=self.inputs, outputs=predictions)
return model
[docs]class ResNet_18_multioutput_3D(ResNet):
[docs] def create_model(self):
self.check_minimum_input_size(self.inputs, [9, 9, 9])
stride_size = self.get_corrected_stride_size(self.inputs, [2, 2, 2], [7, 7, 7])
conv_1 = Conv3D(
filters=64,
kernel_size=[7, 7, 7],
strides=stride_size,
padding="valid",
activation="relu",
)(self.inputs)
stride_size = self.get_corrected_stride_size(conv_1, [2, 2, 2], [3, 3, 3])
pooling_1 = MaxPooling3D(pool_size=[3, 3, 3], strides=stride_size, padding="valid",)(conv_1)
res_block_1 = self.get_residual_identity_block(pooling_1, 64, [3, 3, 3])
res_block_2 = self.get_residual_identity_block(res_block_1, 64, [3, 3, 3])
res_block_2 = self.make_dropout_layer(res_block_2)
res_block_3 = self.get_residual_conv_block(res_block_2, 128, [3, 3, 3])
res_block_4 = self.get_residual_identity_block(res_block_3, 128, [3, 3, 3])
res_block_4 = self.make_dropout_layer(res_block_4)
res_block_5 = self.get_residual_conv_block(res_block_4, 256, [3, 3, 3])
res_block_6 = self.get_residual_identity_block(res_block_5, 256, [3, 3, 3])
res_block_6 = self.make_dropout_layer(res_block_6)
predictions = []
for i_output in self.outputs.values():
res_block_7_output_branch = self.get_residual_conv_block(res_block_6, 512, [3, 3, 3])
res_block_8_output_branch = self.get_residual_identity_block(
res_block_7_output_branch, 512, [3, 3, 3]
)
gap_1_output_branch = GlobalAveragePooling3D()(res_block_8_output_branch)
gap_1_output_branch = self.make_dropout_layer(gap_1_output_branch)
predictions.append(i_output(gap_1_output_branch))
model = Model(inputs=self.inputs, outputs=predictions)
return model
[docs]class ResNet_34_3D(ResNet):
[docs] def create_model(self):
self.check_minimum_input_size(self.inputs, [9, 9, 9])
stride_size = self.get_corrected_stride_size(self.inputs, [2, 2, 2], [7, 7, 7])
conv_1 = Conv3D(
filters=64,
kernel_size=[7, 7, 7],
strides=stride_size,
padding="valid",
activation="relu",
)(self.inputs)
stride_size = self.get_corrected_stride_size(conv_1, [2, 2, 2], [3, 3, 3])
pooling_1 = MaxPooling3D(pool_size=[3, 3, 3], strides=stride_size, padding="valid",)(conv_1)
res_block_1 = self.get_residual_identity_block(pooling_1, 64, [3, 3, 3])
res_block_2 = self.get_residual_identity_block(res_block_1, 64, [3, 3, 3])
res_block_3 = self.get_residual_identity_block(res_block_2, 64, [3, 3, 3])
res_block_3 = self.make_dropout_layer(res_block_3)
res_block_4 = self.get_residual_conv_block(res_block_3, 128, [3, 3, 3])
res_block_5 = self.get_residual_identity_block(res_block_4, 128, [3, 3, 3])
res_block_6 = self.get_residual_identity_block(res_block_5, 128, [3, 3, 3])
res_block_7 = self.get_residual_identity_block(res_block_6, 128, [3, 3, 3])
res_block_7 = self.make_dropout_layer(res_block_7)
res_block_8 = self.get_residual_conv_block(res_block_7, 256, [3, 3, 3])
res_block_9 = self.get_residual_identity_block(res_block_8, 256, [3, 3, 3])
res_block_10 = self.get_residual_identity_block(res_block_9, 256, [3, 3, 3])
res_block_11 = self.get_residual_identity_block(res_block_10, 256, [3, 3, 3])
res_block_12 = self.get_residual_identity_block(res_block_11, 256, [3, 3, 3])
res_block_13 = self.get_residual_identity_block(res_block_12, 256, [3, 3, 3])
res_block_13 = self.make_dropout_layer(res_block_13)
res_block_14 = self.get_residual_conv_block(res_block_13, 512, [3, 3, 3])
res_block_15 = self.get_residual_identity_block(res_block_14, 512, [3, 3, 3])
res_block_16 = self.get_residual_identity_block(res_block_15, 512, [3, 3, 3])
gap_1 = GlobalAveragePooling3D()(res_block_16)
gap_1 = self.make_dropout_layer(gap_1)
predictions = self.outputs(gap_1)
model = Model(inputs=self.inputs, outputs=predictions)
return model