Dynamics-aware deep predictive adaptive scanning optical coherence tomography

Area of Science:

• Biomedical Optics
• Medical Imaging
• Machine Learning in Medicine

Background:

• Conventional optical coherence tomography (OCT) faces a fundamental tradeoff between image resolution and frame rate.
• High-resolution OCT imaging is often limited by slow acquisition speeds, hindering real-time applications.

Purpose of the Study:

• To develop and evaluate machine learning (ML)-based adaptive scanning methods to overcome the OCT frame rate/resolution tradeoff.
• To enhance OCT scanning speed and generalizability for improved clinical and surgical applications.

Main Methods:

• Proposed two ML-based adaptive scanning pipelines utilizing ConvLSTM and temporal attention unit (TAU) models for scene dynamics prediction.
• Integrated ML models with a kinodynamic path planner based on the clustered traveling salesperson problem.
• Validated techniques using deterministic phantoms and real-time surgical tool tracking experiments.

Main Results:

• Achieved mean frame rate speed-ups of up to 40% compared to conventional raster scanning and probabilistic adaptive scanning, without compromising image quality.
• Demonstrated improved generalizability across diverse scenes, reducing the need for manual system parameter tuning.
• Real-time surgical tool tracking showed an average speed-up factor of over 3.2× compared to conventional methods.

Conclusions:

• ML-based adaptive scanning effectively resolves the OCT frame rate/resolution limitation.
• The proposed methods offer enhanced speed, generalizability, and robustness for OCT imaging in various settings, including surgery.