A neural architecture search algorithm for volumetric (3D) segmentation of the pancreas and pancreatic tumor from CT image. This model is trained using the neural network model from the neural architecture search algorithm, DiNTS .
The training dataset is the Pancreas Task from the Medical Segmentation Decathalon. Users can find more details on the datasets at http://medicaldecathlon.com/.
The data list/split can be created with the script
python scripts/prepare_datalist.py --path /path-to-Task07_Pancreas/ --output configs/dataset_0.json
The training was performed with at least 16GB-memory GPUs.
Actual Model Input: 96 x 96 x 96
The neural architecture search was performed with the following:
The training was performed with the following:
The segmentation of pancreas region is formulated as the voxel-wise 3-class classification. Each voxel is predicted as either foreground (pancreas body, tumour) or background. And the model is optimized with gradient descent method minimizing soft dice loss and cross-entropy loss between the predicted mask and ground truth segmentation.
If you face memory issues with CacheDataset, you can either switch to a regular Dataset class or lower the caching rate
cache_rate in the configurations within range [0, 1] to minimize the System RAM requirements.
Dice score is used for evaluating the performance of the model. This model achieves a mean dice score of 0.62.
Please note that this bundle is non-deterministic because of the trilinear interpolation used in the network. Therefore, reproducing the training process may not get exactly the same performance. Please refer to https://pytorch.org/docs/stable/notes/randomness.html#reproducibility for more details about reproducibility.
The loss over 3200 epochs (the bright curve is smoothed, and the dark one is the actual curve)
The mean dice score over 3200 epochs (the bright curve is smoothed, and the dark one is the actual curve)
Users can install Graphviz for visualization of searched architectures (needed in decode_plot.py). The edges between nodes indicate global structure, and numbers next to edges represent different operations in the cell searching space. An example of searched architecture is shown as follows:
In addition to the Pythonic APIs, a few command line interfaces (CLI) are provided to interact with the bundle. The CLI supports flexible use cases, such as overriding configs at runtime and predefining arguments in a file.
For more details usage instructions, visit the MONAI Bundle Configuration Page.
python -m scripts.search run --config_file configs/search.yaml
torchrun --nnodes=1 --nproc_per_node=8 -m scripts.search run --config_file configs/search.yaml
python -m monai.bundle run --config_file configs/train.yaml
Please note that if the default dataset path is not modified with the actual path in the bundle config files, you can also override it by using
python -m monai.bundle run --config_file configs/train.yaml --dataset_dir <actual dataset path>
trainconfig to execute multi-GPU training:
torchrun --nnodes=1 --nproc_per_node=8 -m monai.bundle run --config_file "['configs/train.yaml','configs/multi_gpu_train.yaml']"
trainconfig to execute evaluation with the trained model:
python -m monai.bundle run --config_file "['configs/train.yaml','configs/evaluate.yaml']"
python -m monai.bundle run --config_file configs/inference.yaml
python -m monai.bundle ckpt_export network_def --filepath models/model.ts --ckpt_file models/model.pt --meta_file configs/metadata.json --config_file configs/inference.yaml
 He, Y., Yang, D., Roth, H., Zhao, C. and Xu, D., 2021. Dints: Differentiable neural network topology search for 3d medical image segmentation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (pp. 5841-5850).
Copyright (c) MONAI Consortium
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.
By pulling and using the MONAI Toolkit container and MONAI models, you accept the terms and conditions of the NVIDIA AI Product Agreement. Register for a free evaluation license to try NVIDIA AI Enterprise on your compatible, on-premises system or on the cloud!