Dual-convolutional neural network-enhanced strain estimation method for optical coherence elastography.

Area of Science:

• Biomedical Optics
• Medical Imaging
• Machine Learning

Background:

• Strain estimation is crucial for phase-sensitive optical coherence elastography (PhS-OCE) to assess tissue mechanical properties.
• Current methods may face limitations in accuracy or cost-effectiveness for practical applications.

Purpose of the Study:

• To introduce a novel Dual-Convolutional Neural Network (Dual-CNN) method for improving strain estimation in PhS-OCE.
• To enable high-quality strain estimation using cost-effective, standard-resolution PhS-OCE systems.

Main Methods:

• A Dual-CNN approach was developed, utilizing two PhS-OCE systems: one high-resolution for training data and one standard-resolution for measurements.
• High-resolution strain results and a pre-existing CNN served as training labels, while standard-resolution phase data acted as input.
• The network was trained to estimate high-quality strain from standard-resolution PhS-OCE phase data.

Main Results:

• The Dual-CNN method successfully estimated high-quality strain from standard-resolution PhS-OCE phase results.
• Comparison experiments demonstrated that the Dual-CNN preserved strain estimation quality even with over 80% reduction in light source bandwidth.
• This indicates robustness to reduced optical system specifications.

Conclusions:

• The proposed Dual-CNN method offers a significant advancement in strain estimation for PhS-OCE.
• It allows for accurate and reliable strain measurements using more accessible and cost-effective standard-resolution systems.
• This technique has the potential to broaden the clinical and research applications of PhS-OCE.