High-resolution in vivo 4D-OCT fish-eye imaging using 3D-UNet with multi-level residue decoder

Area of Science:

• Biomedical Imaging
• Medical Technology
• Artificial Intelligence in Medicine

Background:

• Optical Coherence Tomography (OCT) provides high-resolution 3D imaging of biological tissues.
• Acquiring OCT images in vivo is challenging due to motion artifacts from physiological movements and slow frame rates.
• Existing methods struggle to achieve distortion-free volumetric imaging in real-time.

Purpose of the Study:

• To develop a real-time 4D-OCT system for reconstructing near-distortion-free volumetric images.
• To address motion artifacts in OCT imaging using a deep learning-based reconstruction algorithm.
• To enhance the accuracy and efficiency of OCT image reconstruction.

Main Methods:

• Implementation of a real-time 4D-OCT system utilizing a deep learning reconstruction algorithm.
• Collection of undersampled volumetric OCT images at high speed, followed by real-time upsampling using a Convolutional Neural Network (CNN).
• Comparison and refinement of dual-2D- and 3D-UNet architectures, incorporating multi-level information and 16-bit floating-point precision for optimization.

Main Results:

• The refined and optimized 3D-UNet-based network accurately reconstructs tissue structures.
• The system achieves real-time 4D-OCT imaging at a rate exceeding 10 Hz.
• A low root mean square error (RMSE) of approximately 0.03 was achieved, indicating high reconstruction accuracy.

Conclusions:

• The developed deep learning-based 4D-OCT system effectively overcomes motion artifacts.
• The system enables precise, real-time, high-resolution volumetric imaging of biological tissues.
• This advancement holds significant potential for improved in vivo tissue analysis and diagnostics.