CardioSpectrum: Comprehensive Myocardium Motion Analysis with 3D Deep Learning and Geometric Insights (MICCAI 2024)

Official implementation of CardioSpectrum. Visit our project page for more details and updates.

This repository contains the complete optimization and evaluation process for CardioSpectrum. Given a 3D array representing a 3D cardio CT scan during systole, along with segmentations for the left ventricle (LV) and the myocardium, the program generates a synthetic diastole frame and optimizes CardioSpectrum.

CardioSpectrum Architenture:

Architenture

Installation:

pip install --upgrade pip
pip install -r requirements.txt
pip install git+https://github.com/RobinMagnet/pyFM.git

How to Run

Inference

To run inference with the sample data, execute the script:

python infer.py

The output consists of a 3D NumPy array representing the 3D optical flow and a visual representation.

The predicted flow output will be saved at
sample_scan/output/outputs_[date_and_time]/for_drawing/pred_flow.npy
and the visualized 3 sections at sample_scan/output/outputs_[date_and_time]/for_drawing/w_constraints_pred_vs_gt.png

Numerical and visual logs can be found at sample_scan/output/outputs_[date_and_time]/outputs_[method]_training_[date_and_time]/complete_summary

To run the program with your own data, add your files to the sample_scan/data directory and adjust the file paths in configs/general_config_template.json accordingly.

Evaluation

To run evaluation over a single data sample, execute the script:

python eval.py

All calculated evaluation data will be written to sample_scan/output/outputs_[date_and_time]/errors.json

To run evaluation using numerous data samples and/or synthetic data parameters, follow these steps:

Generate different JSON configuration files for all experiments. You can use exp_managing/json_generator.py to automatically do so.

Run

python eval.py --config_path path/to/config

for all config files.

Carefully adjust all paths, sample names, and parameters in exp_managing/analysis/exp_analyzing.py and exp_managing/analysis/utils/exp_paths.json, then run
```
python exp_managing/analysis/exp_analyzing.py
```
The output is a dataframe collecting all metrics along with plots representing different errors.

The code was tested on a single Quadro RTX 8000 GPU, using CUDA version 10.2 and Python version 3.8.0.

The provided sample scan is the cardio CT scan originally from 3D Slicer sample data.

Citing

If you find this repository helpful, please cite our publication.

arXiv preprint:

@misc{zuler2024cardiospectrumcomprehensivemyocardiummotion,
      title={CardioSpectrum: Comprehensive Myocardium Motion Analysis with 3D Deep Learning and Geometric Insights}, 
      author={Shahar Zuler and Shai Tejman-Yarden and Dan Raviv},
      year={2024},
      eprint={2407.03794},
      archivePrefix={arXiv},
      primaryClass={eess.IV},
      url={https://arxiv.org/abs/2407.03794}, 
}

MICCAI 2024 proceedings:

COMING SOON

shaharzuler / CardioSpectrum

readme