AI-driven quantification of ground glass opacities in lungs of COVID-19 patients using 3D computed tomography imaging Paper Link

Data

We used "MosMedData: Chest CT Scans with COVID-19 Related Findings" (https://mosmed.ai/datasets/covid19_1110). Data needs to download before using this pipeline. The size of the data is 33 GB. All the data is in .nii format. Hence, you need to convert them into .png for using this pipeline. You can use nii_to_pngs_converter.py for this purpose.

Lung and GGOs Segmentation

This is the first step of this pipeline. All the relevant codes have been stored inside the folder "Lung_GGO_segmentation"

1_original_to_binary.py first script of this pipeline.

This code has been used for generating binary masks from original images. data_path is the original data path. The output of this code will be saved at internal_path.

1a_convex_hull_CT_monjoy.py second script of this pipeline

This code has been used for getting convex hull, convex points. The input is binary images. data_path is a path of dataset. The results will be saved at save_path.

2_Original_lung_from_binary.py third script of this pipeline

This code can be used to extract the only lung from the original images and binary mask (resulted from step-1 code). original_image_path is an original image path. mask_image_path is a path of masks. All the results will be saved at save_path.

2a_remove_white_regions_.py fourth script of this pipeline

This code removes white non-GGO regions from the original segmented images. original_image_path is an original image path. original_mask_path is a path of masks. All the results will be saved at save_path.

3_K-means_final.py fifth script of this pipeline

3_K-means_final.py is used to segment GGOs. original_image_path is the original image path. OrgMask_path is the mask path. All the results will be saved at save_path.

Point Cloud

2a_convert_monjoy.py first script of this pipeline

To convert stack of images into point cloud use 2a_convert_monjoy.py. The code is stored inside the folder 1_Convert_images_to_point_cloud. In our study, we used sixteen consecutive images. You can use as many images as you want. There are no restrictions. Set image path image_path and save results at save_path. All the final files will be saved in .ply format.

visualize_pcd_file.py second script of this pipeline

This script can be used to visualize point clouds.

downsample_main_code_24dec.py third script of this pipeline

To downsample the number of points of the point cloud, you can use downsample_main_code_24dec.py. The code is stored inside the folder 2_Downsample_pointCloud. ply_file_path denotes .ply file path and downsampled_files_save_path represents the path, where downsampled files will be saved. In our study, we downsampled the number of points from 800K to 2480 points.

3_Ply_to_pcd_monjoy fourth script of this pipeline

This script is used for converting from .ply format to .pcd format. .ply and .pcd both files represent point clouds. For our easy analysis, we converted all .ply to .pcd. The script is stored inside the folder 3_Ply_to_pcd_monjoy.

Convert .pcd to .txt format fifth script of this pipeline

For the PointNet++ data needs to be converted .txt format. Hence, we used a script inside 4_prepare_data_for_pointNet/pointcloudToTXT-master/build/ folder. Use the below command for this purpose. cd /build For .pcd files ./pointcloudToTXT <pcd file> -o <output dir> For .ply files ./pointcloudToTXT <ply file> -o <output dir>

PointNet++

PointNet++ related scripts are saved inside PointNet++ folder. For model training run train.py and for testing run test.py. MOre details on PointNet++ can be found at https://github.com/charlesq34/pointnet2.