A pre-trained model for 3D segmentation of the spleen organ from CT images using DeepEdit.
DeepEdit is an algorithm that combines the power of two models in one single architecture. It allows the user to perform inference as a standard segmentation method (i.e., UNet) and interactively segment part of an image using clicks . DeepEdit aims to facilitate the user experience and, at the same time, develop new active learning techniques.
The model was trained on 32 images and validated on 9 images.
The training dataset is the Spleen Task from the Medical Segmentation Decathalon. Users can find more details on the datasets at http://medicaldecathlon.com/.
The training as performed with the following:
Dice score is used for evaluating the performance of the model. This model achieves a dice score of 0.97, depending on the number of simulated clicks.
spleen_deepedit_annotation bundle supports acceleration with TensorRT through the ONNX-TensorRT method. The table below displays the speedup ratios observed on an A100 80G GPU.
|method||torch_fp32(ms)||torch_amp(ms)||trt_fp32(ms)||trt_fp16(ms)||speedup amp||speedup fp32||speedup fp16||amp vs fp16|
model computationmeans the speedup ratio of model's inference with a random input without preprocessing and postprocessing
end2endmeans run the bundle end-to-end with the TensorRT based model.
torch_ampare for the PyTorch models with or without
trt_fp16are for the TensorRT based models converted in corresponding precision.
speedup fp16are the speedup ratios of corresponding models versus the PyTorch float32 model
amp vs fp16is the speedup ratio between the PyTorch amp model and the TensorRT float16 based model.
Currently, the only available method to accelerate this model is through ONNX-TensorRT. However, the Torch-TensorRT method is under development and will be available in the near future.
This result is benchmarked under:
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.
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 monai.bundle run --config_file configs/train.json
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.json --dataset_dir <actual dataset path>
trainconfig to execute multi-GPU training:
torchrun --standalone --nnodes=1 --nproc_per_node=2 -m monai.bundle run --config_file "['configs/train.json','configs/multi_gpu_train.json']"
Please note that the distributed training-related options depend on the actual running environment; thus, users may need to remove
--nnodes, or do some other necessary changes according to the machine used. For more details, please refer to pytorch's official tutorial.
trainconfig to execute evaluation with the trained model:
python -m monai.bundle run --config_file "['configs/train.json','configs/evaluate.json']"
python -m monai.bundle run --config_file configs/inference.json
python -m monai.bundle trt_export --net_id network_def \ --filepath models/model_trt.ts --ckpt_file models/model.pt \ --meta_file configs/metadata.json --config_file configs/inference.json \ --precision <fp32/fp16> --use_onnx "True" --use_trace "True"
python -m monai.bundle run --config_file "['configs/inference.json', 'configs/inference_trt.json']"
 Diaz-Pinto, Andres, et al. DeepEdit: Deep Editable Learning for Interactive Segmentation of 3D Medical Images. MICCAI Workshop on Data Augmentation, Labelling, and Imperfections. MICCAI 2022.
 Diaz-Pinto, Andres, et al. "MONAI Label: A framework for AI-assisted Interactive Labeling of 3D Medical Images." arXiv preprint arXiv:2203.12362 (2022).
 Sakinis, Tomas, et al. "Interactive segmentation of medical images through fully convolutional neural networks." arXiv preprint arXiv:1903.08205 (2019).
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!