Facilitating Radiograph Interpretation: Refined Generative Models for Precise Bone Suppression in Chest X-rays

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Chest X-rays (CXRs) are vital for diagnosing lung diseases, but bone structures obscure a significant percentage of lung cancers.
• Current computer-aided detection (CAD) systems for bone suppression face challenges including inefficiency, poor generalization, and information loss.

Purpose of the Study:

• To develop a novel, efficient, and accurate end-to-end architecture for automated bone suppression in CXRs.
• To improve the detection of lung nodules by reducing interference from bone structures.

Main Methods:

• Introduced a mask-guided model leveraging the Pix2Pix framework for enhanced computational efficiency (92.5% parameter reduction).
• Incorporated a rib mask-guided module with a mask encoder and cross-attention mechanism for spatial constraints and reduced information loss.
• Utilized digitally reconstructed radiographs (DRRs) for initial training and fine-tuned on the JSRT dataset.

Main Results:

• The mask-guided model demonstrated superior bone suppression performance compared to state-of-the-art methods.
• Achieved high performance metrics on the JSRT dataset: SSIM of 0.99 ± 0.002 and PSNR of 36.14 ± 1.13.
• Showcased significant improvements in processing speed and reduced model size.

Conclusions:

• The proposed mask-guided model effectively suppresses bone structures in CXRs, enhancing diagnostic accuracy for lung nodules.
• The model's efficiency and accuracy make it suitable for deployment on affordable, low-power devices in diverse clinical settings.
• This advancement supports earlier clinical decision-making and improves lung cancer detection rates.