Robust, real-time, digital focusing for FD-OCM using ISAM on a GPU

摘要

Frequency domain optical coherence tomography (FD-OCT) achieves high image acquisition speeds by probing all depths of a sample simultaneously. However, the tightly focused beam required for frequency domain optical coherence microscopy (FD-OCM) produces images with poor lateral resolution at depths away from the beam waist. The new technique of interferometric synthetic aperture microscopy (ISAM) can digitally focus these poorly resolved FD-OCM images, resulting in uniform lateral resolution throughout the sample volume equivalent to that in the plane of focus of the incident beam. While ISAM is computationally intensive, we demonstrate that an ISAM implementation using Nvidia's parallel Compute Unified Device Architecture (CUDA) can achieve real-time focusing using a mid-range Nvidia GPU. The time required for digital focusing scales linearly with image size, at a rate of about 10 nanoseconds per voxel. This makes possible real-time FD-OCM. For example, a 3-D image (512×512×128 voxels) with cross-section 1.2 mm × 1.2 mm and 200 micron depth requires 17 seconds to acquire with a 100 kHz A-scan rate (and 6 repeated x-scans for motion sensitivity), but only 360 milliseconds to focus with ISAM. This example image is simulated with a numerical aperture (NA) of 0.07, so that the 200 micron depth represents four Rayleigh ranges (± 2 Rayleigh ranges from the focal plane). In addition, our simulations indicate that ISAM performs well with very noisy input data. Even with noise levels as high as 50%, ISAM produces focused images with signal-to-noise ratios of over 100. ISAM-focusing is both fast and robust.