Accurate attenuation characterization in optical coherence tomography using multi-reference phantoms and deep...

Area of Science:

• Biomedical Optics
• Medical Imaging
• Optical Coherence Tomography

Background:

• The optical attenuation coefficient (AC) is vital for quantitative tissue analysis and differentiation.
• Accurate AC quantification from optical coherence tomography (OCT) signals is a significant challenge in clinical applications.

Purpose of the Study:

• To investigate factors affecting AC extraction accuracy in existing OCT algorithms.
• To develop a novel deep learning approach (MR-Net) for improved AC quantification from OCT data.

Main Methods:

• A Multi-Reference Phantom Driven Network (MR-Net) was developed using deep learning and multi-reference phantoms.
• Intralipid and silicone-TiO2 phantoms with known AC values were imaged using a 1300 nm swept-source OCT system.
• Performance was compared against existing OCT-based AC extraction algorithms, analyzing data length, out-of-focus distance, and phantom properties.

Main Results:

• MR-Net demonstrated superior performance across all evaluated metrics.
• MR-Net achieved an average relative error of 10.43% for AC calculation, significantly outperforming existing methods (lowest 23.72%).
• The network effectively models complex physical effects influencing OCT signal propagation.

Conclusions:

• MR-Net offers a highly accurate and automated method for quantifying the optical attenuation coefficient from OCT signals.
• This approach provides a robust quantitative framework for disease diagnosis and enhanced tissue characterization in clinical settings.