Real-time 4D signal processing and visualization using graphics processing unit on a regular nonlinear-k Fourier-domain OCT system

Abstract

We realized graphics processing unit (GPU) based real-time 4D (3D+time) signal processing and visualization on a regular Fourier-domain optical coherence tomography (FD-OCT) system with a nonlinear k-space spectrometer. An ultra-high speed linear spline interpolation (LSI) method for lambda-to-k spectral re-sampling is implemented in the GPU architecture, which gives average interpolation speeds of >3,000,000 line/s for 1024-pixel OCT (1024-OCT) and >1,400,000 line/s for 2048-pixel OCT (2048-OCT). The complete FD-OCT signal processing including lambda-to-k spectral re-sampling, fast Fourier transform (FFT) and post-FFT processing have all been implemented on a GPU. The maximum complete A-scan processing speeds are investigated to be 680,000 line/s for 1024-OCT and 320,000 line/s for 2048-OCT, which correspond to 1GByte processing bandwidth. In our experiment, a 2048-pixel CMOS camera running up to 70 kHz is used as an acquisition device. Therefore the actual imaging speed is camera- limited to 128,000 line/s for 1024-OCT or 70,000 line/s for 2048-OCT. 3D Data sets are continuously acquired in real time at 1024-OCT mode, immediately processed and visualized as high as 10 volumes/second (12,500 A-scans/volume) by either en face slice extraction or ray-casting based volume rendering from 3D texture mapped in graphics memory. For standard FD-OCT systems, a GPU is the only additional hardware needed to realize this improvement and no optical modification is needed. This technique is highly cost-effective and can be easily integrated into most ultrahigh speed FD-OCT systems to overcome the 3D data processing and visualization bottlenecks.

Fig. 1

System configuration; CMOS, CMOS line scan camera; L, spectrometer lens; G, reflective grating; C1, C2, achromatic collimators; C, 50:50 broadband fiber coupler; CL, camera link cable; COMP, host computer; GPU, graphics processing unit; PCIE-X16, PCI Express x16 2.0 interface; MON, Monitor; GVS, galvanometer mirror pairs; R1, R2, relay lens; SL, scanning lens; RG, reference glass; SP, Sample.

Fig. 2

CPU-GPU hybrid system architecture.

Fig. 3

Flowchart of parallelized LSI. Blue blocks: memory for pre-stored data; yellow blocks: memory for real-timely refreshed data.

Fig. 4

(a) Schematic of ray-casting CPU-GPU hybrid architecture; (b) flowchart of interactive volume rendering by GPU.

Fig. 5

(a) GPU processing time versus one-batch A-scan number; (b) GPU processing line rate versus one-batch A-scan number.

Fig. 6

System sensitivity roll-off: (a) 1024-OCT; (b) 2048-OCT.

Fig. 7

B-scan images of an infrared sensing card: (a) 1024-OCT, 10,000 A-scan/frame, 12.8 frame/s; (b) 2048-OCT, 10,000 A-scan/frame, 7.0 frame/s. The scale bars represent 250µm in both dimensions.

Fig. 8

En face slices reconstructed from real-timely acquired and processed volumetric data, the scale bar represents 100µm for all images: (a) 250 × 160 × 512 voxels; (b) from the same volume as (a) but 25 µm deeper; (c) 250 × 80 × 512 voxels; (d) from the same volume as (c) but 25 µm deeper; (e) 125 × 80 × 512 voxels; (f) from the same volume as (e) but 25 µm deeper.

Fig. 9

(a) ( Media 1) The dynamic 3D OCT movie of a piece of sugar-shell coated chocolate; (b) sugar-shell top truncated by the X-Y plane, inner structure visible; (c) a five-layer phantom.

Fig. 10

In vivo real-time 3D imaging of a human finger tip. (a) ( Media 2) Skin and fingernail connection; (b) ( Media 3) Fingerprint, side-view with “L” volume rendering frame; (c) ( Media 4) Fingerprint, top-view.

Fig. 11

( Media 5) Multiple 2D frames real-time rendering from the same 3D data set with different model view matrix.