Author: Fernández Hernández, Alberto
Date: 2022 - 07 - 13
The main purpose is create a model to automatically segment the stomach and intestines (small and large) on MRI scans, in order to outline the position of the stomach and intestines to adjust the direction of the x-ray beams to increase the dose delivery to the tumor and avoid main organs. A method to segment the stomach and intestines would make treatments much faster and would allow more patients to get more effective treatment. The MRI scans are from actual cancer patients who had 1-5 MRI scans on separate days during their radiation treatment.
Two models are proposed:
- UNet VS Feature Pyramid Network (FPN)
| # Model | Number of parameters | Backbone | Inference Time (GPU) - minutes * |
|---|---|---|---|
| # Unet | 8.7 M | Efficientnet-B1 | 2:50 min. |
| # FPN | 8.2 M | Efficientnet-B1 | 3:06 min. |
* 3759 non-empty-masks images are "inferred", with batch size: 1
Non empty masks
| # Model | Dice score large bowel | Dice score small bowel | Dice score stomach |
|---|---|---|---|
| # Unet | 0.81 | 0.79 | 0.90 |
| # FPN | 0.73 | 0.73 | 0.89 |
Empty masks
| # Model | Dice score large bowel | Dice score small bowel | Dice score stomach |
|---|---|---|---|
| # Unet | 0.99 | 0.99 | 0.99 |
| # FPN | 0.95 | 0.95 | 0.99 |
UW-Madison GI Tract Image Segmentation - Kaggle dataset
- Storage: Google Cloud Storage
- Code: Python 3.7 + Google Cloud functions
- Libraries:
- PyTorch
- Albumentations
- Image Segmentation Models (smp) library
- Pydicom
- Sci-kit Learn
- Matplotlib
- Streamlit
- Skimage
- tqdm
- wandb
- OpenCV
- Numpy
- Google-cloud
UW-Madison GI Tract Image Segmentation - Kaggle dataset
Segmentation models PyTorch libraries
A prior knowledge guided deep learning based semi-automatic segmentation for complex anatomy on MRI
A2-FPN for Semantic Segmentation of Fine-Resolution Remotely Sensed Images