Scoliosis_project

This is an official repo for the paper on automatic scoliosis angle measurement: Deep learning automates Cobb angle measurement compared with multi-expert observers.

Provided by: Hanxue Gu

Installation

1. Create and Activate a Conda Environment

conda create -n spine-env python=3.8 -y
conda activate spine-env

conda install pytorch==1.7.0 torchvision==0.8.1 cudatoolkit=11.0 -c pytorch

# With custom CUDA ops
pip install mmdet==2.16.0
pip install mmcv-full==1.3.16

# OR, basic version (no custom CUDA ops):
# pip install mmdet==2.16.0
# pip install mmcv==1.3.16

pip install pycocotools matplotlib scikit-image jupyterlab

2. Verify your installation

python -c "
import torch, torchvision, mmcv, mmdet
print('Torch:', torch.__version__)
print('TorchVision:', torchvision.__version__)
print('MMCV:', mmcv.__version__)
print('MMDetection:', mmdet.__version__)

Data Preparation

1. Inference: Place your sample image (in PNG format) under ./example_case.
2. Training/Validation: Organize your dataset into ./train and ./val folders, and provide additional JSON files that reference these images. Examples of the JSON format can be found in:

Examples of json format can be found in:

raw_train_label_new.json
raw_val_label_new.json

Train/Validate the Model

Use the notebook train_and_val.ipynb for model training and validation.

Inference/Test on a Single Case with Visualization

Use the notebook inference_1_case.ipynb to perform inference on a single case. The results, including the main Cobb angle annotation, will be displayed step by step.

Model weights

The model weights can be found through: Model weights, and you can put them under /work_dirs.

Comments

We built a detection algorithm mainly based on MMDetection. The details of MMDetection can be found here. If you find our repo helpful, please feel free to cite our work:

@misc{li2024deeplearningautomatescobb,
      title={Deep learning automates Cobb angle measurement compared with multi-expert observers}, 
      author={Keyu Li and Hanxue Gu and Roy Colglazier and Robert Lark and Elizabeth Hubbard and Robert French and Denise Smith and Jikai Zhang and Erin McCrum and Anthony Catanzano and Joseph Cao and Leah Waldman and Maciej A. Mazurowski and Benjamin Alman},
      year={2024},
      eprint={2403.12115},
      archivePrefix={arXiv},
      primaryClass={eess.IV},
      url={https://arxiv.org/abs/2403.12115}, 
}