CMT-AMAI24paper

Paper

Quantifying Knee Cartilage Shape and Lesion: From Image to Metrics

AMAI’24 (MICCAI workshop) (in press)

TL;DR

CMT, a toolbox for knee MRI analysis, model training, and visualization.

Contributions

Joint Template-Learning and Registration Mode – CMT-reg
CartiMorph Toolbox (CMT)

Models for CMT

models (both for segmentation and registration) for this work – can be loaded into CMT
more models from the CMT models page

Quick Start

For model evaluation and the training of other SoTA models:

git clone https://github.com/YongchengYAO/CMT-AMAI24paper.git
cd CMT-AMAI24paper
conda create --name CMT_AMAI24paper --file env.txt

For model training in CMT, see the instructions for CMT

Code

We compared the proposed CMT-reg with other template learning and/or registration models – Aladdin and LapIRN.

Code for Aladdin training, inference, and evaluation (for reproducing the results in Tables 2 & 3)
Code for LapIRN training, inference, and evaluation (for reproducing the results in Tables 2 & 3)
Code for CMT evaluation (for reproducing the results in Table 3)
Training, inference, and evaluation of CMT-reg are implemented in CMT, set these parameters in CMT:
- Cropped Image Size: 64, 128, 128
- Training Epoch: 2000
- Network Width: x3
- Loss: MSE+LNCC

Data

This is the data used for reproducing Tables 2 & 3.

Data for this repo for model training, inference, and evaluation

# data folder structure
├── Code
  ├── Aladdin
    ├── Model
  ├── LapIRN
    ├── Model
├── Data
  ├── Aladdin
  ├── CMT_data4AMAI 
  ├── LapIRN

How to use files in the Data folder?
1. clone this repo: CMT-AMAI24paper
2. put the Data folder under CMT-AMAI24paper/
How to use files in the Code folder?
1. clone this repo: CMT-AMAI24paper
2. put corresponding Model folders to
  - CMT-AMAI24paper/Code/Aladdin/
  - CMT-AMAI24paper/Code/LapIRN/

Raw Data

MR Image: OAI
Annotation: OAI-ZIB

Data Information: here (link CMT-ID to OAI-SubjectID)

Processed Data from CartiMorph

This is the data used for training CMT-reg and nnUNet in CMT

If you use the processed data, please note that the manual segmentation annotations come from this work:

Automated Segmentation of Knee Bone and Cartilage combining Statistical Shape Knowledge and Convolutional Neural Networks: Data from the Osteoarthritis Initiative (https://doi.org/10.1016/j.media.2018.11.009)

Citation

(conference proceedings in press)

@misc{yao2024quantifyingkneecartilageshape,
      title={Quantifying Knee Cartilage Shape and Lesion: From Image to Metrics}, 
      author={Yongcheng Yao and Weitian Chen},
      year={2024},
      eprint={2409.07361},
      archivePrefix={arXiv},
      primaryClass={eess.IV},
      url={https://arxiv.org/abs/2409.07361}, 
}

Acknowledgment

The training, inference, and evaluation code for Aladdin and LapIRN are adapted from these GitHub repos:

Aladdin: https://github.com/uncbiag/Aladdin
LapIRN: https://github.com/cwmok/LapIRN

CMT is based on CartiMorph: https://github.com/YongchengYAO/CartiMorph

@article{YAO2024103035,
title = {CartiMorph: A framework for automated knee articular cartilage morphometrics},
journal = {Medical Image Analysis},
author = {Yongcheng Yao and Junru Zhong and Liping Zhang and Sheheryar Khan and Weitian Chen},
volume = {91},
pages = {103035},
year = {2024},
issn = {1361-8415},
doi = {https://doi.org/10.1016/j.media.2023.103035}
}