How to predict a CT scan without nodules annotation

[lokeycookie]

Hello all!

Let’s say that I have succesfully trained a 3D unet model and have saved its weight. Currently in RadIO tutorial 4, in order to predict a CT scan, we first generate cancer and noncancer crops using split_dump function. Afterwards, we use combine_crops to generate a batch and then the model will predict it. Current workflow is as such:

However, in order for the crops to generate, we first need to obtain the nodules location/annotations.

Now, if we have a CT scan with no nodule annotations, how do I use the model to predict if there is any nodules in the CT scan? I ask this as currently, we need to have nodules annotation before doing 3D cropping ( via sample_nodules function). So how do we process a CT scan with no nodule annotations, crop it to volume size of (32,64,64) such that the model is able to make a prediction?

PS: The model input shape is (None,1,32,64,64)

Did anyone able to resolve this?