Deep Learning for High Speed Optical Coherence Elastography With a Fiber Scanning Endoscope.

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Soft Tissue Mechanics

Background:

• Tissue stiffness is crucial for diagnosing soft tissue pathologies.
• Current clinical imaging methods for tissue stiffness are often unsuitable for intraoperative use.
• Minimally invasive surgery requires real-time, localized mechanical property assessment.

Purpose of the Study:

• To develop a miniaturized fiber scanning endoscope for real-time elastography.
• To implement a deep learning (DL) pipeline for processing complex wave field data.
• To enable fast and localized elasticity estimation during interventions.

Main Methods:

• A miniaturized fiber scanning endoscope with optimized probe design for conical scan patterns (5.05kHz temporal frequency).
• A deep learning-based spatio-temporal network trained end-to-end on phantom data for elasticity estimation.
• Imaging of complex, diffuse wave fields across multiple frequencies and directions.

Main Results:

• The DL approach achieved significantly lower mean absolute errors in elasticity estimation compared to conventional phase tracking (4.48kPa vs 19.75kPa in 3D).
• Real-time elasticity mapping was demonstrated with localized and robust estimates.
• Feasibility was shown in ex-vivo porcine tissue.

Conclusions:

• The developed fiber scanning endoscope and DL pipeline offer a promising solution for real-time intraoperative tissue elastography.
• This technology can improve diagnostic accuracy and guidance during minimally invasive procedures.
• Further validation in clinical settings is warranted.