3D Spine Shape Estimation from Single 2D DXA (MICCAI2024 Oral)

Table of Contents

1. Project Description
2. Getting Started
   • Prerequisites
3. Usage
4. Acknowledgments

About The Project

In this project, we aim to understand the 3D geometry of the spine at a patient-specific level given a single 2D DXA scan. For instance, scoliosis involves deformations in multiple directions with known deformations on the coronal, sagittal and axial plane. A better understanding of the deformations will lead to tailored treatments and better management of patients. We propose an automated general framework to estimate the 3D spine shape from 2D DXA scans. We achieve this by predicting the coronal view (centerline and lateral curves) as well as sagittal view (centerline and lateral curves). Given 2 orthogonal projections of the 3D spine, we can reconstruct the 3D spine. This gives a user-friendly visualisation of the spine that could be used in complement to 2D DXA to measure scoliosis.

2 main contributions :

• Regression of 3D patient-specific spine shapes from 2D AP DXA only.
• Use of lightweight transformer and ResNet50 backbone surpassing Transformer and CNN-based models

Getting Started

Please follow the below instructions to run the code.

Prerequisites

1. Download the paired DXA-MRI data used in this work

From the UKBiobank, this dataset can be downloaded after creating an account and registering on the UKBiobank platform. Follow instructions from this repo: https://github.com/rwindsor1/UKBiobankDXAMRIPreprocessing

2. Install the required packages

pip install -r requirements.txt

Usage

To train the model for spine curves regression, run

python train.py

Can now be trained on a single GPU 10GB VRAM! Training time ~6hrs

You can download pre-trained model checkpoints [here] (https://www.dropbox.com/scl/fi/be4dg1xccgl1fo9wn74i8/epoch-996-loss_valid-points-best_loss-0.0168.pt?rlkey=ytnrrctofyebqtkj5p4554px1&st=zujv8ure&dl=0) and place in the checkpoints folder.

At inference time to obtain orthogonal spine curves from DXA scan, run

python test.py

ToDo

• Training code
• Testing code
• Documentation
• Demo coming soon

Acknowledgments

If you found this work useful, please cite the following papers:

@InProceedings{Bou_3D_MICCAI2024,
        author = { Bourigault, Emmanuelle and Jamaludin, Amir and Zisserman, Andrew},
        title = { { 3D Spine Shape Estimation from Single 2D DXA } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
        year = {2024},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15005},
        month = {October},
        page = {pending}
}

@inproceedings{Windsor21,
   author    = {Rhydian Windsor and Amir Jamaludin and Timor Kadir and Andrew Zisserman},
   booktitle = {Proc. Medical Image Computing and Computer Aided Intervention (MICCAI)},
   title     = {Self-Supervised Multi-Modal Alignment for Whole Body Medical Imaging},
   date      = {2021},
 }