High-Resolution Chest X-Ray Bone Suppression Using Unpaired CT Structural Priors.

Area of Science:

• Medical Imaging
• Radiology
• Computer-Aided Diagnosis

Background:

• Bone structures in Chest X-rays (CXRs) can obscure lung abnormalities, impacting diagnostic accuracy.
• Bone-free CXRs are not always readily available, necessitating advanced suppression techniques.

Purpose of the Study:

• To develop and validate a coarse-to-fine CXR bone suppression method utilizing structural priors from unpaired computed tomography (CT) images.
• To improve the diagnostic value of CXRs for both human radiologists and computer-aided diagnosis systems.

Main Methods:

• A coarse-to-fine approach using digitally reconstructed radiographs (DRRs) derived from CT as an intermediary.
• Domain adaptation techniques and Laplacian of Gaussian (LoG) filtering to bridge domain differences between CXR and DRR.
• Upsampling and subtraction of decomposed bone structures from original CXRs to achieve bone suppression.

Main Results:

• The proposed method surpasses existing unsupervised CXR bone suppression techniques.
• Bone-suppressed CXRs significantly reduced the false-negative rate for lung disease detection by radiologists from 15% to 8%.
• Deep learning models utilizing bone-suppressed CXRs achieved state-of-the-art disease classification performance.

Conclusions:

• The developed CXR bone suppression technique effectively removes bone structures, enhancing diagnostic utility.
• This method offers a valuable tool for improving the accuracy of lung disease diagnosis from CXRs.
• The approach demonstrates potential for advancing computer-aided diagnosis in medical imaging.