4D deep learning for real-time volumetric optical coherence elastography

Area of Science:

• Biomedical Optics
• Medical Imaging
• Machine Learning

Background:

• Soft tissue elasticity is crucial for disease diagnosis and treatment.
• Shear wave velocity is a common method for estimating tissue stiffness.
• Optical coherence elastography (OCE) offers high spatial and temporal resolution but is often slow for clinical use due to sequential data acquisition.

Purpose of the Study:

• To develop a faster, more practical approach for optical coherence elastography.
• To enable direct volumetric elastography estimations from phase image data.

Main Methods:

• Utilized a fast imaging device to acquire small image volumes at 831 Hz.
• Employed a 4D convolutional neural network for processing spatial and temporal phase image data.
• Evaluated the method on gelatin phantoms with known elasticity.

Main Results:

• The neural network accurately predicted gelatin concentration in unseen samples with a mean error of 0.65 ± 0.81 percentage points.
• Achieved rapid data acquisition (under 12 ms) and processing (under 22 ms).
• Demonstrated direct volumetric optical coherence elastography from phase image data.

Conclusions:

• The proposed method allows for direct volumetric optical coherence elastography.
• It bypasses the need for specific stimulation or sampling sequences.
• Enables estimation of elastic tissue properties at frequencies up to 40 Hz, enhancing clinical applicability.