Chest X-Ray Diagnostic Quality Assessment: How Much Is Pixel-Wise Supervision Needed?

Area of Science:

• Medical Imaging
• Radiology
• Computer Vision

Background:

• Chest X-rays are vital for diagnosing chest diseases.
• Poor radiograph quality negatively impacts diagnosis and increases patient burden.
• Existing algorithms and datasets for chest X-ray quality assessment are lacking.

Purpose of the Study:

• To develop an effective algorithm for chest radiograph diagnostic quality assessment.
• To analyze image characteristics of common quality issues (Scapula Overlapping Lung, Artifact, Lung Field Loss, Clavicle Unflatness).
• To create a public dataset for chest X-ray quality assessment.

Main Methods:

• Analysis of image characteristics for four key quality issues.
• Proposed a multi-label semantic segmentation framework with a distance map estimation.
• Applied weakly supervised learning for segmentation to reduce annotation dependency.
• Introduced the ChestX-rayQuality dataset with detailed annotations.

Main Results:

• General image classification methods are insufficient for quality assessment.
• The proposed semantic segmentation approach significantly improves quality assessment accuracy.
• Weakly supervised segmentation achieved performance close to fully supervised methods.
• The ChestX-rayQuality dataset provides valuable resources for research.

Conclusions:

• A novel semantic segmentation-based method effectively assesses chest X-ray diagnostic quality.
• The developed method addresses limitations of traditional approaches and deep learning models.
• Weak supervision offers a viable alternative for annotation-intensive tasks.
• The publicly released dataset will facilitate future research and development in this area.