CheXmask: a large-scale dataset of anatomical segmentation masks for multi-center chest x-ray images

TL;DR

This work introduces an extensive chest X-ray multi-center segmentation dataset with uniform and fine-grain anatomical annotations for images coming from five well-known publicly available databases, resulting in 657,566 segmentation masks.

Abstract

The development of successful artificial intelligence models for chest X-ray analysis relies on large, diverse datasets with high-quality annotations. While several databases of chest X-ray images have been released, most include disease diagnosis labels but lack detailed pixel-level anatomical segmentation labels. To address this gap, we introduce an extensive chest X-ray multi-center segmentation dataset with uniform and fine-grain anatomical annotations for images coming from five well-known publicly available databases: ChestX-ray8, Chexpert, MIMIC-CXR-JPG, Padchest, and VinDr-CXR, resulting in 657,566 segmentation masks. Our methodology utilizes the HybridGNet model to ensure consistent and high-quality segmentations across all datasets. Rigorous validation, including expert physician evaluation and automatic quality control, was conducted to validate the resulting masks. Additionally, we provide individualized quality indices per mask and an overall quality estimation per dataset. This dataset serves as a valuable resource for the broader scientific community, streamlining the development and assessment of innovative methodologies in chest X-ray analysis. The CheXmask dataset is publicly available at: https://physionet.org/content/chexmask-cxr-segmentation-data/

Figures (10)

• Figure 1: Data processing flowchart depicting the main steps involved in the building of the CheXmask dataset.
• Figure 2: Summary diagram of inclusion-exclusion criteria.
• Figure 3: Histogram showcasing the distribution of the RCA-estimated DSC for the complete CheXmask database. Example images with their landmark-based segmentations were drawn with lines to their corresponding histogram bin. Reproduction of x-ray images was allowed upon request to the original sources.
• Figure 4: Label Studio setup for manual landmark-based segmentation. Reproduction of x-ray images was allowed upon request to the original sources.
• Figure 5: Illustration of the Mean Squared Error (MSE) per landmark across the entire gold-standard set per expert reviewer, depicted in logarithmic scale for improved visual clarity. The color intensity of each landmark represents the magnitude of the MSE. Moreover, the size of each landmark is proportional to its respective MSE, thus larger landmarks indicate a greater prediction error.