BS-LDM: Effective Bone Suppression in High-Resolution Chest X-Ray Images With Conditional Latent Diffusion Models

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Radiology

Background:

• Chest X-rays (CXRs) are crucial for diagnosing lung diseases but are limited by overlapping bone structures.
• Bone interference in CXRs can lead to misdiagnosis of pulmonary lesions.
• Existing methods struggle with effective bone suppression while preserving image details.

Purpose of the Study:

• To develop an advanced framework, BS-LDM, for high-resolution Chest X-ray (CXR) bone suppression.
• To improve the accuracy and reliability of pulmonary lesion detection in CXRs.
• To enhance the clinical utility of CXR imaging through improved image quality.

Main Methods:

• Developed an end-to-end framework, BS-LDM, utilizing conditional latent diffusion models.
• Incorporated a multi-level hybrid loss-constrained vector-quantized generative adversarial network for perceptual compression.
• Employed offset noise in the forward process and temporal adaptive thresholding in the reverse process for enhanced generation.
• Created the SZCH-X-Rays dataset and processed JSRT dataset images for bone suppression tasks.

Main Results:

• BS-LDM demonstrated superior performance in bone suppression on high-resolution CXR images.
• The framework effectively preserved important details in soft tissue images.
• Downstream evaluations confirmed the clinical value and effectiveness of BS-LDM.
• The developed SZCH-X-Rays dataset provides a valuable resource for research.

Conclusions:

• BS-LDM offers a significant advancement in medical image processing for lung disease diagnosis.
• The framework's ability to suppress bone structures enhances the detection of pulmonary abnormalities.
• BS-LDM holds considerable clinical value for radiologists and healthcare professionals.
• The study provides an open-source implementation for further research and development.