def __init__(self):
super(PetExtract, self).__init__()
in_channels = 1
self.conv_1 = nn.Conv3d(in_channels=in_channels,
out_channels=32,
kernel_size=3,
padding=1)
self.resnet_1 = _make_block(ResNet, input_channels=32)
def __init__(self):
super(AttnPet, self).__init__()
self.upsample_1 = _make_block(UpSample, input_channels=256)
self.resnet_10 = _make_block(ResNet, input_channels=128)
self.gate_1 = PetAttentionBlock3D(in_channels=128,
pet_channels=32,
gating_channels=256,
inter_channels=128)
self.upsample_2 = _make_block(UpSample, input_channels=128)
self.resnet_11 = _make_block(ResNet, input_channels=64)
self.gate_2 = PetAttentionBlock3D(in_channels=64,
pet_channels=32,
gating_channels=256,
inter_channels=64)
self.upsample_3 = _make_block(UpSample, input_channels=64)
self.resnet_12 = _make_block(ResNet, input_channels=32)
self.gate_3 = PetAttentionBlock3D(in_channels=32,
pet_channels=32,
gating_channels=256,
inter_channels=32)
self.conv_2 = nn.Conv3d(in_channels=32,
out_channels=1,
kernel_size=3,
padding=1)
def __init__(self):
super(Reg, self).__init__()
self.conv_1 = nn.Conv3d(in_channels=1, out_channels=32, kernel_size=3, padding=1)
self.drop = nn.Dropout(p=0.2)
self.resnet_1 = _make_block(ResNet, input_channels=32)
self.downsample_1 = _make_block(DownSample, input_channels=32)
self.resnet_2 = _make_block(ResNet, input_channels=64)
self.resnet_3 = _make_block(ResNet, input_channels=64)
self.downsample_2 = _make_block(DownSample, input_channels=64)
self.resnet_4 = _make_block(ResNet, input_channels=128)
self.resnet_5 = _make_block(ResNet, input_channels=128)
self.downsample_3 = _make_block(DownSample, input_channels=128)
self.resnet_6 = _make_block(ResNet, input_channels=256)
self.resnet_7 = _make_block(ResNet, input_channels=256)
self.resnet_8 = _make_block(ResNet, input_channels=256)
self.resnet_9 = _make_block(ResNet, input_channels=256)
self.reg = _make_block(Regularizer)
def __init__(self, mask=None):
super(SegPET, self).__init__()
self.mask = mask
self.conv_1 = nn.Conv3d(in_channels=1,
out_channels=32,
kernel_size=3,
padding=1)
self.drop = nn.Dropout(p=0.2)
self.resnet_1 = _make_block(ResNet, input_channels=32)
self.downsample_1 = _make_block(DownSample, input_channels=32)
self.resnet_2 = _make_block(ResNet, input_channels=64)
self.resnet_3 = _make_block(ResNet, input_channels=64)
self.downsample_2 = _make_block(DownSample, input_channels=64)
self.resnet_4 = _make_block(ResNet, input_channels=128)
self.resnet_5 = _make_block(ResNet, input_channels=128)
self.downsample_3 = _make_block(DownSample, input_channels=128)
self.resnet_6 = _make_block(ResNet, input_channels=256)
self.resnet_7 = _make_block(ResNet, input_channels=256)
self.resnet_8 = _make_block(ResNet, input_channels=256)
self.resnet_9 = _make_block(ResNet, input_channels=256)
self.seg = Segmentation()
self.pet = PET()
self.grid_layer = None
if self.mask == 'mask_prf':
self.grid_layer = GridAttentionBlock3D(in_channels=1,
gating_channels=256,
inter_channels=1)
def __init__(self, both=False, attention=False):
super(Seg, self).__init__()
if both:
in_channels = 2
else:
in_channels = 1
self.conv_1 = nn.Conv3d(in_channels=in_channels,
out_channels=32,
kernel_size=3,
padding=1)
self.drop = nn.Dropout(p=0.2)
self.resnet_1 = _make_block(ResNet, input_channels=32)
self.downsample_1 = _make_block(DownSample, input_channels=32)
self.resnet_2 = _make_block(ResNet, input_channels=64)
self.resnet_3 = _make_block(ResNet, input_channels=64)
self.downsample_2 = _make_block(DownSample, input_channels=64)
self.resnet_4 = _make_block(ResNet, input_channels=128)
self.resnet_5 = _make_block(ResNet, input_channels=128)
self.downsample_3 = _make_block(DownSample, input_channels=128)
self.resnet_6 = _make_block(ResNet, input_channels=256)
self.resnet_7 = _make_block(ResNet, input_channels=256)
self.resnet_8 = _make_block(ResNet, input_channels=256)
self.resnet_9 = _make_block(ResNet, input_channels=256)
if attention:
self.seg = AttnSeg()
else:
self.seg = Segmentation()
def __init__(self):
super(SegAttnPet, self).__init__()
in_channels = 1
self.conv_1 = nn.Conv3d(in_channels=in_channels,
out_channels=32,
kernel_size=3,
padding=1)
self.drop = nn.Dropout(p=0.2)
self.resnet_1 = _make_block(ResNet, input_channels=32)
self.downsample_1 = _make_block(DownSample, input_channels=32)
self.resnet_2 = _make_block(ResNet, input_channels=64)
self.resnet_3 = _make_block(ResNet, input_channels=64)
self.downsample_2 = _make_block(DownSample, input_channels=64)
self.resnet_4 = _make_block(ResNet, input_channels=128)
self.resnet_5 = _make_block(ResNet, input_channels=128)
self.downsample_3 = _make_block(DownSample, input_channels=128)
self.resnet_6 = _make_block(ResNet, input_channels=256)
self.resnet_7 = _make_block(ResNet, input_channels=256)
self.resnet_8 = _make_block(ResNet, input_channels=256)
self.resnet_9 = _make_block(ResNet, input_channels=256)
self.pet_extract = PetExtract()
self.seg = AttnPet()