Learning-based distortion correction enables proximal-scanning endoscopic OCT elastography.

Area of Science:

• Biomedical Engineering
• Optical Imaging
• Medical Devices

Background:

• Proximal scanning in endoscopic and intravascular Optical Coherence Tomography (OCT) is cost-effective but suffers from non-uniform rotational distortion (NURD).
• NURD significantly limits the application of OCT in functional imaging techniques like OCT elastography (OCE).

Purpose of the Study:

• To develop and validate a learning-based method for accurate NURD correction in proximal-scanning OCT systems.
• To enable stable imaging for intensity-based OCE, facilitating advanced vascular assessments.

Main Methods:

• Extraction of real distortion vectors from an endoscopic OCT system for model training.
• Application of both convolutional neural network (CNN) and transformer-based learning architectures for NURD correction.
• Validation of elastography calculations using digital image correlation and optical flow.

Main Results:

• The proposed method, utilizing real distortion vectors, demonstrated superior accuracy in NURD correction compared to methods using pseudo vectors.
• The corrected OCT images enabled effective 2D displacement and strain distribution analysis in both radial and circumferential directions.
• Proximal-scanning endoscopic OCE successfully differentiated varying stiffness in atherosclerotic vascular phantoms.

Conclusions:

• Learning-based NURD correction using real distortion vectors is effective for enabling stable endoscopic OCE.
• The developed method overcomes limitations of previous approaches and enhances the capabilities of OCT for vascular imaging.
• This advancement allows for comprehensive 2D mechanical property assessment of atherosclerotic plaques.