chexmask / DATA_DICTIONARY.md
ming0100's picture
Upload DATA_DICTIONARY.md with huggingface_hub
41de2d4 verified
|
Raw
History Blame Contribute Delete
2.39 kB

CheXmask Database Data Dictionary

CSV File Structure

Each dataset in the CheXmask Database is provided as a separate CSV file. Below are detailed descriptions of all fields/variables present in these files.

Image Identification

Field Name: Image ID
Description: Reference identifier linking to original image in source dataset
Type: String
Notes: Format varies by source dataset (ChestX-ray8, CheXpert, MIMIC-CXR-JPG, Padchest, VinDr-CXR)

Quality Metrics

Field Name: Dice RCA (Max)
Description: Maximum Dice Similarity Coefficient from Reverse Classification Accuracy
Type: Float
Range: 0.0 to 1.0
Units: Dimensionless
Notes: Higher values indicate better segmentation quality

Field Name: Dice RCA (Mean)
Description: Mean Dice Similarity Coefficient from Reverse Classification Accuracy
Type: Float
Range: 0.0 to 1.0
Units: Dimensionless
Notes: Recommended threshold for use is >= 0.7

Anatomical Features

Field Name: Landmarks
Description: Set of points representing organ contours generated by HybridGNet model
Type: Array of coordinates
Format: JSON array of [x,y] coordinates
Units: Pixels

Field Name: Left Lung
Description: Segmentation mask for left lung
Type: String
Format: Run-length encoding (RLE)
Notes: Must be decoded using provided dimensions

Field Name: Right Lung
Description: Segmentation mask for right lung
Type: String
Format: Run-length encoding (RLE)
Notes: Must be decoded using provided dimensions

Field Name: Heart
Description: Segmentation mask for heart
Type: String
Format: Run-length encoding (RLE)
Notes: Must be decoded using provided dimensions

Image Dimensions

Field Name: Height
Description: Height of segmentation mask
Type: Integer
Units: Pixels
Notes: Required for decoding RLE masks

Field Name: Width
Description: Width of segmentation mask
Type: Integer
Units: Pixels
Notes: Required for decoding RLE masks

Additional Notes

  1. RLE Format: Run-length encoding is used to compress the binary mask data. Each RLE string represents pairs of (start_position, run_length) for the mask. A decoding script is present in the GitHub repository.