QOCT-Net: A Physics-Informed Neural Network for Intravascular Optical Coherence Tomography Attenuation Imaging

Area of Science:

• Biomedical Optics
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Intravascular optical coherence tomography (IVOCT) offers high-resolution imaging of coronary arteries.
• Accurate quantitative attenuation imaging is crucial for characterizing tissue and identifying vulnerable plaques.

Purpose of the Study:

• To develop a deep learning-based method for quantitative IVOCT attenuation imaging.
• To enable precise tissue characterization and vulnerable plaque detection using IVOCT data.

Main Methods:

• Proposed a physics-informed deep network, Quantitative OCT Network (QOCT-Net).
• QOCT-Net recovers pixel-level optical attenuation coefficients from IVOCT B-scan images.
• The method is based on the multiple scattering model of light transport.

Main Results:

• QOCT-Net demonstrated superior visual and quantitative attenuation coefficient estimates.
• Significant improvements were observed in structural similarity (≥7%), energy error depth (≥5%), and peak signal-to-noise ratio (≥12.4%) compared to existing methods.
• The network was validated on both simulated and in vivo datasets.

Conclusions:

• The developed deep learning method enables high-precision quantitative IVOCT imaging.
• This approach holds potential for improved tissue characterization and vulnerable plaque identification in clinical settings.