Integrated deep learning framework for accelerated optical coherence tomography angiography.

Area of Science:

• Biomedical Imaging
• Optical Coherence Tomography
• Deep Learning

Background:

• Label-free optical coherence tomography angiography (OCTA) is crucial for microvasculature imaging.
• Current OCTA methods suffer from slow imaging speeds due to high sampling density and multiple B-scan acquisitions.
• Existing deep learning (DL) approaches partially address acquisition speed limitations.

Purpose of the Study:

• To develop an integrated DL method to accelerate OCTA acquisition speed and enhance image quality.
• To address both undersampling and repeated B-scan requirements simultaneously.
• To improve reconstruction performance beyond current DL-based methods.

Main Methods:

• An end-to-end deep neural network (DNN) framework was designed.
• A two-staged adversarial training scheme was employed for reconstruction.
• The DNN reconstructs high-quality, fully-sampled angiograms from undersampled, low-quality data in two stages: enhancing resolution and then image quality.

Main Results:

• The proposed DL framework achieved superior reconstruction performance on an in-vivo mouse brain vasculature dataset.
• Quantitative and qualitative assessments demonstrated enhanced speed and quality compared to conventional methods.
• OCTA imaging speed was accelerated from 16 to 256 frames per second while preserving image quality.

Conclusions:

• The integrated DL method offers a significant advancement in OCTA imaging speed and quality.
• This software-only solution enables faster and more efficient preclinical and clinical studies.
• The developed DNN framework provides a practical approach to overcome OCTA's inherent speed limitations.