This dataset was be modified from Lung segmentation dataset by Kónya et al., 2020
https://www.kaggle.com/sandorkonya/ct-lung-heart-trachea-segmentation
The original nrrd files were re-saved in single tensor format with masks corresponding to labels: (lungs, heart, trachea) as numpy arrays using pickle.
Each tensor has the following shape:
number of slices, width, height, number of classes
Where:
- the width and height number of slices are individual parameters of each tensor id
- number of classes = 3
In addition, the data was re-saved as RGB images, where each image corresponds to one ID slice, and their mask-images have channels corresponding to three classes: (lung, heart, trachea).
The dataset contains:
- images.zip
- images in RGB format
- masks.zip
- segmentation masks in RGB format
- train.csv
- csv file with image names