GPU-accelerated iterative method for FD-OCT image reconstruction with an image-level cross-domain regularizer.

Area of Science:

• Biomedical Optics
• Medical Imaging Technology
• Computational Imaging

Background:

• Traditional Fourier-domain optical coherence tomography (FD-OCT) image reconstruction relies on inverse discrete Fourier transform (IDFT) or slow A-line-based iterative methods.
• A-line-based iterative methods are computationally intensive and incompatible with existing image-level processing techniques, limiting their practical application.

Purpose of the Study:

• To develop a novel, efficient, and versatile iterative image reconstruction method for FD-OCT.
• To enable simultaneous image processing during reconstruction for enhanced image quality.
• To demonstrate the method's scalability and potential for real-time 3D OCT imaging.

Main Methods:

• Proposed a B-scan-based iterative reconstruction approach for FD-OCT.
• Implemented GPU acceleration for large-scale parallelism and reduced computational time.
• Developed a novel image-level cross-domain regularizer for integrated image processing during reconstruction.

Main Results:

• Achieved significantly higher image quality (improved SNR and CNR) compared to state-of-the-art iterative methods and conventional workflows.
• Reduced computational time by orders of magnitude, enabling real-time reconstruction at over 20 B-scans per second.
• Demonstrated scalability for 3D OCT image reconstruction and successful integration of image enhancement.

Conclusions:

• The proposed B-scan-based iterative method offers superior performance in terms of image quality and speed for FD-OCT.
• GPU acceleration and integrated image-level processing provide substantial advantages over existing techniques.
• The method holds significant potential for real-world applications requiring high-speed, high-quality OCT imaging.