A pre-trained model for volumetric (3D) segmentation of lung tumors from CT images.

This model is trained using the runnerup [1] awarded pipeline of the “Medical Segmentation Decathlon Challenge 2018” using the AHnet architecture [2] with 50 training images and 13 validation images.


For more detailed description of “Medical Segmentation Decathlon Challenge 2018,” please see

The training dataset is Task06_lung.tar from the link above.

The data must be converted to 1mm resolution before training:


tlt-dataconvert -d ${SOURCE_IMAGE_ROOT} -r 1 -s .nii.gz -e .nii.gz -o ${DESTINATION_IMAGE_ROOT}

NOTE: to match up with the default setting, we suggest that ${DESTINATION_IMAGE_ROOT} match DATA_ROOT as defined in environment.json in this MMAR’s config folder.

Training configuration

The training was performed with command, which required 12GB-memory GPUs.

Training Graph Input Shape: dynamic

Actual Model Input: 96 x 96 x 96

Input and output formats


Single channel CT image


Two channels

  • Label 1: lung

  • Label 0: everything else


This Dice scores on the validation data achieved by this model are:

  1. lung: 0.417

In order to access this model, please apply for general availability access at

This model is usable only as part of Transfer Learning & Annotation Tools in Clara Train SDK container. You can download the model from NGC registry as described in Getting Started Guide.

The content of this model is only an example. It is not intended to be a substitute for professional medical advice, diagnosis, or treatment.

[1] Xia, Yingda, et al. “3D Semi-Supervised Learning with Uncertainty-Aware Multi-View Co-Training.” arXiv preprint arXiv:1811.12506 (2018).

[2] Liu, Siqi, et al. “3d anisotropic hybrid network: Transferring convolutional features from 2d images to 3d anisotropic volumes.” International Conference on Medical Image Computing and Computer-Assisted Intervention. Springer, Cham, 2018.

© Copyright 2018-2020, NVIDIA Corporation. All rights reserved. Last updated on Feb 1, 2023.