• CadiacSeg3D.MindSpore
• Content
• • 3DUNet Descriptions
  • Model Architecture
  • Dataset
  • Environment Requirements
  • Quick Start
  • Script Description
  • Script Parameters
  • Training Process
  • Evaluation Process
  • • Evaluation
  • Model Description
  • • Performance
    • • Training Performance
      • Inference Performance

CadiacSeg3D.MindSpore

Content

• 3DUNet Descriptions
• Model Architecture
• Dataset
• Environment Requirements
• Quick Start
• Script Description
• Script Parameters
• Training Process
• Evaluation Process
  • Evaluation
• Model Description
  • Performance
    • Training Performance
    • Inference Performance

3DUNet Descriptions

3DUNet was proposed in 2016, it is a type of neural network that directly consumes volumetric images. The 3DUNet extends the previous u-net architecture by replacing all 2D operations with their 3D counterparts. The implementation performs on-the-fly elastic deformations for efficient data augmentation during training. It is trained end-to-end from scratch, i.e., no pre-trained network is required.

Paper: Çiçek, Özgün, et al. "3D U-Net: learning dense volumetric segmentation from sparse annotation." International conference on medical image computing and computer-assisted intervention. Springer, Cham, 2016.

Model Architecture

The 3DUNet segementation network takes n 3D volumetric images as input, applies input and feature transformations. BN is introdued before each ReLU.

Dataset

Dataset used: MM-WHS, 3d-Covid-segmentation.

MM-WHS

• Dataset size: 1.2G
  • Train: 428M, 20 images and corresponding labels.
  • Test: 761M
• Data format: NifTI files
  • Notes: Data will be processed in src/dataset.py

Covid-19 CT image dataset （3D）

Dataset size: 3.5G

• Train: 2.8G, 160 images and corresponding labels.
• Test: 724M, 49 images and corresponding labels.

Environment Requirements

• Hardware（Ascend）
  • Prepare hardware environment with Ascend processor.
• Framework
  • MindSpore
• For more information, please check the resources below：
  • MindSpore Tutorials
  • MindSpore Python API

• suwen package

  pip install -r requirements.txt
  pip install ./suwen-1.0.1-py3-none-any.whl
  

Quick Start

After installing MindSpore via the official website, you can start training and evaluation as follows:

# enter script dir, train PointNet
sh run_train_ascend.sh

# enter script dir, evaluate PointNet
sh run_eval.sh

Script Description

.
└── 3dUNet
    ├── README.md
    ├── UNet3d.ckpt
    ├── eval.py
    ├── eval_log.txt
    ├── requirements.txt
    ├── scripts
    │   ├── run_eval.sh
    │   └── run_train_ascend.sh
    ├── src
    │   ├── __init__.py
    │   ├── config.py
    │   ├── convert_nifti.py
    │   ├── dataset.py
    │   ├── dice_metric.py
    │   ├── loss.py
    │   ├── lr_schedule.py
    │   ├── transform.py
    │   └── utils.py
    ├── suwen-1.0.1-py3-none-any.whl
    ├── train.py
    └── train_log.txt

Script Parameters

Major parameters in train.py are as follows:

--data_path: The absolute full path to the train and evaluation datasets.
--seg_path : The absolute full path to the train and evaluation segmentation labels.
--ckpt_path: The absolute full path to the checkpoint file saved after training.

More hyperparamteters can be modified in src/config.py.

Training Process

• running on Ascend

  sh run_train_ascend.sh
  

  After training, the loss value will be achieved as what in train_log.txt

  The model checkpoint will be saved in the current ckpt directory.

Evaluation Process

Evaluation

Before running the command below, please check the checkpoint path used for evaluation.

• running on Ascend

  sh scripts/run_eval.sh
  

  You can view the results through the file "eval_log". The accuracy of the test dataset will be as what in eval_log.txt.

Model Description

Performance

Training Performance

Parameters
Resource	Ascend 910; CPU 2.60GHz, 24cores; Memory, 96G
uploaded Date	11/15/2021 (month/day/year)
MindSpore Version	1.3.0
Dataset	MM-WHS
Training Parameters	epoch=600, steps=, batch_size = , lr=
Optimizer	Adam
Loss Function	Softmax Cross Entropy
outputs	probability
Loss	SoftmaxCrossEntropyWithLogits
Speed	212.469 ms/step-
Total time	3399.497 ms
Checkpoint for Fine tuning	56M (.ckpt file)

Inference Performance

Parameters
Resource	Ascend 910; CPU 2.60GHz, 24cores; Memory, 96G
uploaded Date	05/29/2021 (month/day/year)
MindSpore Version	1.3.0
Dataset	MM-WHS
batch_size	1
outputs	probability
Dice	85.08%