Convolutional neural network-based image enhancement for x-ray percutaneous coronary intervention

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Cardiovascular Interventions

Background:

• Percutaneous coronary intervention (PCI) relies on X-ray imaging, posing risks of high radiation dose and contrast media-induced nephropathy.
• Reducing X-ray and contrast doses leads to lower image quality (noise, reduced contrast), potentially compromising diagnostic accuracy.
• Existing denoising methods may not adequately preserve crucial vessel wall details in PCI images.

Purpose of the Study:

• To develop and evaluate a vessel-edge-preserving convolutional neural network (CNN) for enhancing low-dose percutaneous coronary intervention (PCI) images.
• To optimize CNN objective functions for effective denoising while preserving vessel wall integrity.
• To compare the proposed CNN method against established denoising techniques.

Main Methods:

• Simulated low-dose PCI images were created by adding artificial noise to high-dose images.
• Vessel-edge-preserving CNNs were designed and their objective functions optimized for denoising and contrast enhancement.
• The CNN-based method was quantitatively and qualitatively evaluated and compared with K-SVD and block-matching and 3D filtering.

Main Results:

• The proposed CNN-based method demonstrated promising performance in enhancing simulated low-dose PCI images.
• The optimized CNN objective functions effectively preserved vessel wall edges.
• The CNN approach showed superior or comparable results to K-SVD and block-matching and 3D filtering in denoising and enhancement.

Conclusions:

• CNNs offer a powerful tool for real-time denoising and contrast enhancement in PCI imaging.
• The developed CNN-based method can facilitate safer PCI procedures by enabling the use of lower radiation and contrast media doses.
• This approach holds potential for improving diagnostic accuracy and patient safety in interventional cardiology.