A-scan sequence transformers for palpation with optical coherence elastography.

Area of Science:

• Biomedical Optics
• Medical Imaging
• Machine Learning

Background:

• Accurate tissue elasticity recognition aids surgical navigation, but palpation is subjective and limited in minimally invasive surgery.
• High-speed optical coherence elastography (OCE) offers intraoperative elasticity estimation via mechanical wave propagation.
• Robust wave velocity estimation and elastic modulus reconstruction are challenging due to wave propagation modeling complexities.

Purpose of the Study:

• To develop and evaluate deep learning models for end-to-end elasticity estimation directly from OCE phase data.
• To investigate the suitability of transformer architectures for processing OCE A-scan sequences.
• To compare the performance of deep learning methods against conventional techniques and CNN-based approaches.

Main Methods:

• Utilized transformer-based deep learning models to process temporal sequences of 1D axial scans (A-scans) from OCE.
• Trained and tested models on homogeneous tissue phantoms with known elastic properties.
• Validated generalization on heterogeneous phantoms and assessed elasticity of biological soft tissues (heart, kidney, liver).

Main Results:

• Achieved a mean error of 1.64 kPa for elasticity estimation in homogeneous phantoms.
• Demonstrated significant improvement over conventional processing (7.80 kPa error) and CNN-based methods (5.55 kPa error).
• Successfully generalized to heterogeneous phantoms and accurately assessed elasticity in various soft tissue samples.

Conclusions:

• Transformer architectures are highly effective for reconstructing tissue elasticity from OCE A-scan sequences.
• Deep learning, particularly transformers, offers a robust and accurate solution for intraoperative elasticity estimation.
• This approach has the potential to enhance surgical navigation and tissue characterization.