Predict Readmission of HFpEF patients using LV strains from cine CMR

Author: Zhennong Chen, PhD

This is the GitHub repo based on an paper under rewivew:
Predicting Two-year Readmission of Heart Failure with Preserved Ejection Fraction using Comprehensive Left-ventricular Strain Analysis from Cardiac Magnetic Resonance and Deep Learning
Authors: Zhennong Chen, Hui Ren, Chagin Choi, Hyun Jin Cho, Sun-hyun Kim, Siyeop Yoon, Xiang Li, Quanzheng Li

Citation: TBD

Description

We have proposed to use LV strain maps derived from cine CMR with a graph convolutional network (GCN) to predict heart failure (HF)-related readmission in patients with heart failure with preserved ejection fraction (HFpEF).
The main contributions are as follows:
(1) LV strain map is a [N,T] matrix where N is the number of AHA segments (regional) and T is the number of time frames (temporal). It is a comprehensive representation of LV function.
(2) A Chebyshev GCN is used to encode the LV strain map. The edge matrix is defined according to the anatomical adjacency of AHA segments.
(3) We also offer the option to add clinical data from electronic health record (EHR) as additional input.

User Guideline

Environment Setup

The entire code is containerized. This makes setting up environment swift and easy. Make sure you have nvidia-docker and Docker CE installed on your machine before going further.

• You can build your own docker from the folder docker. The docker image can be built by ./docker_build.sh, after that the docker container can be built by ./docker_run.sh. The installed packages can be referred to dockerfile and requirements.txt
  
• You'll need docker/docker_tensorflow for step 1 and docker/docker_torch for the step 2
  

Data Preparation (we have examples available)

You should prepare two things before running this step. Please refer to the example_data folder for guidance:

• NIfTI images of cine CMR data and corresponding segmentation (myocardium + LV bloodpool).

  • Please re-organize your data so that each time frame (3D volume) is a separate nii file.
  • The segmentations were done by our cineCMR segmentation foundation model, or you can prepare by your own.
  • These are saved in example_data/raw_nii_images

• A patient list that enumerates all your cases.

  • To understand the expected format, please refer to the file:
    example_data/Patient_lists/patient_list.xlsx.
  • The ground truth label of re-admission is also provided in this patient list

Experiments

we have designed our study into 2 steps, with each step having its own jupyter notebook.
step1: get LV strain map: step1_calculate_LV_strain.ipynb

1. We use published method DeepStrain to generate regional LV strain from cine CMR short-axis (SAX) views. we use the first time frame (end-diastole) as template.
   
2. The generated strains are saved in example_data/results

step2: model*: use step2_model.ipynb

1. It interpolates the number of time frames into 25 as default
   
2. Training and prediction included
   

Additional guidelines

Please contact chenzhennong@gmail.com for any further questions.