Issue training a custom segmentation model in MONAI Label using 3D Slicer and AWS

[arbizu148613]

Hi everyone!

I'm trying to create a new segmentation model in MONAI using 3D Slicer. The goal is to segment the lumbar muscles. I’ve already completed the manual segmentations, and now I want to train the model using an AWS server.

What I’ve done so far is download the Radiology application and the Segmentation model, and modify the segmentation.py script by specifying my own labels and setting use_pretrained_model=False, as explained in the "Training from scratch" video.

My aim is to perform an 80-20 training using 100 MRI studies. To do this, I prepared a dataset with 79 images and their corresponding labels, all organized in the correct folder structure. Then, I planned to load one last image and its segmentation from Slicer, send it to the server, and click Train to start training the full model.

However, at this point the model does not train correctly. When I try to run inference on new cases, it fails to produce an automatic segmentation and throws an error.

Does anyone know what could be causing this? I'm not sure if I have all the necessary files in place or if there’s something else I need to change. Also, I haven’t renamed the model—I’ve kept the default name segmentation, which might also be causing issues.

I hope someone can help! @diazandr3s If you could take a look, it would be greatly appreciated :)
Thanks!