High speed optical coherence tomography (OCT) systems with A-scan rates greater than 100 kHz allow for 4D visualizations in applications such as intraoperative OCT. However, traditional triangle or sawtooth waveforms used to drive galvanometer scanners often have frequency content that exceeds the bandwidth of the scanners, leading to distorted scans. Sinusoidal waveforms used to drive resonant scanners also lead to distorted scans due to the nonlinear scan velocity. Additionally, with raster scan patterns, the scanner needs time to stop and reverse direction in between B-scans, leading to significant acquisition dead time. Continuous scan patterns such as constant frequency spiral scanning or Lissajous scanning no longer have acquisition dead times, but suffer from non-uniform sampling across the imaging plane. We previously introduced constant linear velocity (CLV) spiral scanning as a novel scan pattern to maximize the data acquisition efficiency of high speed OCT systems. While this continuous scan pattern has no acquisition dead time and produces more uniform sampling compared to raster scanning, it required significant processing time. We introduce a processing pipeline implemented using CUDA in C++, which drastically reduces the amount of processing time needed, allowing real time visualization of 4D OCT data. To demonstrate its potential utility, we used CLV scanning with a 100 kHz swept-source OCT system to image retinas of enucleated porcine eyes undergoing mock ophthalmic surgery movements. Additionally, we rendered these volumes in virtual reality (VR) in real time, allowing for interactive manipulation and sectioning.
|