How to use ResNet API?

[markvvw]

Recently, I attempted to use ResNet for building a classification model of 3D MRI (25625616). However, I do not know how to set the parameters as a medical doctor. There were no correct information for addressing this problem.

Thank you in advance for your reply.

FALSE:
model = ResNet(layers=[18], spatial_dims=3,num_classes=2, n_input_channels=1,block='basic',block_inplanes=[2, 2, 2, 2]).to(device)#

[KumoLiu]

Hi @markvvw, you can try resnet18, resnet50 or other resnet in this file, just replace num_classes.

from monai.networks.nets import resnet18
model = resnet18(num_classes=3)

Hope it can help you, thanks!

[KumoLiu]

Hi @markvvw, you could change the input_channel by setting n_input_channels. Thanks!

[Wangrui-berry]

Hello, if I want to use 3d resnet, should I change spatial_dims as 3 to change resnet (conv2d) to 3d?
model = monai.networks.nets.resnet10(spatial_dims=3, n_input_channels=1, n_classes=1)
How can I use conv3d?

[Wangrui-berry]

model = resnet18(num_classes=3)
This code is still use conv2d to process 3d data, isn't it?

[KumoLiu]

Hi @Sudayberry, yes you need to change spatial_dims to 3 if you want to use 3d Resnet but don't need to change conv by yourself. The network will automatically use 3dconv after changing spatial_dims.

Hope it can help you, thanks!

[Wangrui-berry]

Thanks. That's a good answer to my question.