A Deep Learning Method for Motion Artifact Correction in Intravascular Photoacoustic Image Sequence.

Area of Science:

• Biomedical Imaging
• Artificial Intelligence in Medicine
• Cardiovascular Imaging

Background:

• Intravascular photoacoustic (IVPA) imaging is crucial for coronary artery assessment.
• Motion artifacts from cardiac cycles significantly degrade IVPA image quality.
• Current gating methods can lead to loss of valuable diagnostic information.

Purpose of the Study:

• To develop a deep learning-based method for motion artifact correction in non-gated IVPA imaging.
• To preserve diagnostically valuable information lost in traditional gating techniques.
• To validate the method's effectiveness in in vivo intracoronary imaging.

Main Methods:

• A deep learning network, MAC-Net, was designed to correct motion artifacts in IVPA sequences.
• Raw signal frames were clustered into dynamic and static categories.
• The network was trained and tested using a computer-generated dataset.
• The method was validated using in vivo intravascular ultrasound and optical coherence tomography data.

Main Results:

• The MAC-Net successfully corrected motion artifacts in dynamic frames without discarding data.
• Quantitative evaluation showed improved visual quality and reduced inter-frame dissimilarity.
• The method demonstrated motion suppression comparable to gating and image registration techniques.
• Feasibility was confirmed in in vivo intracoronary imaging scenarios.

Conclusions:

• Deep learning offers a powerful solution for motion artifact correction in non-gated IVPA imaging.
• The proposed method enhances image quality and preserves data integrity for better coronary artery assessment.
• This technique holds promise for improving intracoronary imaging diagnostics.