Distortion and instability compensation with deep learning for rotational scanning endoscopic optical coherence...

Area of Science:

• Medical Imaging
• Computational Imaging
• Biomedical Engineering

Background:

• Endoscopic Optical Coherence Tomography (OCT) offers high-speed, high-resolution imaging for minimally invasive procedures.
• Image distortion and instability in endoscopic OCT systems arise from motor irregularities and synchronization issues.
• Real-time artifact correction is crucial for reliable diagnostic and therapeutic applications.

Purpose of the Study:

• To develop and validate a new online correction method for stabilizing endoscopic OCT videos.
• To address B-scan distortion, video stream shaking, and drift caused by A-line shifting.
• To enhance image quality for real-time clinical use.

Main Methods:

• A computational approach integrating Convolutional Neural Networks (CNNs) for azimuthal shifting estimation.
• A dual-CNN architecture to estimate A-line shifting and dynamic orientation angles.
• Training the CNN model using semi-synthetic OCT videos with introduced rotational distortions.

Main Results:

• The proposed method effectively stabilizes real endoscopic OCT videos.
• The CNN model trained on semi-synthetic data demonstrates generalization to real-world artifacts.
• Successful correction of significant scanning artifacts in both ex vivo and in vivo datasets.

Conclusions:

• The developed online correction method significantly improves the stability and quality of endoscopic OCT imaging.
• The CNN-based approach provides a robust solution for real-time artifact compensation in endoscopic OCT.
• This technique has the potential to enhance the utility of OCT in clinical diagnosis and treatment.