File size: 2,695 Bytes
cd6dcce 0bba136 cd6dcce |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 |
# Description
A pre-trained model for inferencing volumetric (3D) kidney substructures segmentation from contrast-enhanced CT images (Arterial/Portal Venous Phase).
A tutorial and release of model for kidney cortex, medulla and collecting system segmentation.
Authors: Yinchi Zhou ([email protected]) | Xin Yu ([email protected]) | Yucheng Tang ([email protected]) |
# Model Overview
A pre-trained UNEST base model [1] for volumetric (3D) renal structures segmentation using dynamic contrast enhanced arterial or venous phase CT images.
## Data
The training data is from the [ImageVU RenalSeg dataset] from Vanderbilt University and Vanderbilt University Medical Center.
(The training data is not public available yet).
- Target: Renal Cortex | Medulla | Pelvis Collecting System
- Task: Segmentation
- Modality: CT (Artrial | Venous phase)
- Size: 96 3D volumes
The data and segmentation demonstration is as follow:
 <br>
## Method and Network
The UNEST model is a 3D hierarchical transformer-based semgnetation network.
Details of the architecture:
 <br>
## Training configuration
The training was performed with at least one 16GB-memory GPU.
Actual Model Input: 96 x 96 x 96
## Input and output formats
Input: 1 channel CT image
Output: 4: 0:Background, 1:Renal Cortex, 2:Medulla, 3:Pelvicalyceal System
## Performance
A graph showing the validation mean Dice for 5000 epochs.
 <br>
This model achieves the following Dice score on the validation data (our own split from the training dataset):
Mean Valdiation Dice = 0.8523
Note that mean dice is computed in the original spacing of the input data.
## commands example
Download trained checkpoint model to ./model/model.pt:
Add scripts component: To run the workflow with customized components, PYTHONPATH should be revised to include the path to the customized component:
```
export PYTHONPATH=$PYTHONPATH:"'<path to the bundle root dir>/scripts'"
```
Execute inference:
```
python -m monai.bundle run evaluating --meta_file configs/metadata.json --config_file configs/inference.json --logging_file configs/logging.conf
```
## More examples output
 <br>
# Disclaimer
This is an example, not to be used for diagnostic purposes.
# References
[1] Yu, Xin, Yinchi Zhou, Yucheng Tang et al. "Characterizing Renal Structures with 3D Block Aggregate Transformers." arXiv preprint arXiv:2203.02430 (2022). https://arxiv.org/pdf/2203.02430.pdf
[2] Zizhao Zhang et al. "Nested Hierarchical Transformer: Towards Accurate, Data-Efficient and Interpretable Visual Understanding." AAAI Conference on Artificial Intelligence (AAAI) 2022
|