DL-CTA coronary Agatston calcium scoring

This repository contains the PyTorch implementation of "Automatic Calcium Scoring in Coronary CT Angiography using Deep Learning: Automatically Derived using Spectral CT and Validated using Multiple CTA Imaging Protocols"

Setup

Required packages

• PyTorch 1.7
• SimpleITK
• numpy

Data Preparation

To train and predict on CTA cases, organize the data according to the following format:

data
├── Images
|   ├── case1_iso.nii.gz
|   └── case2_iso.nii.gz
├── Labels
|   ├── case1_cal_seg.nii.gz
|   ├── case1_cal_map.nii.gz
|   ├── case2_cal_seg.nii.gz
|   └── case2_cal_map.nii.gz
├── train_ID_lis.txt
└── val_ID_lis.txt

where caseID_cal_seg.nii.gz is the calcification segmentation and caseID_cal_map.nii.gz is the CAC score distributution according to voxel-wise calcification severity.

Training

To train the segmentation model, run:

python train_seg.py

To train the CAC score regression model, run:

python train_regress.py

Testing

To run testing on unseen data, first generate the segmentation results by running:

predict_segmentation.py

To run the regression model, run:

predict_regress.py

Note that the segmentation results must be made available before running the regression model.

Results: