[Legacy Report] Parallel GPGPU computations for optical coherence tomography (OTC)
#GPU
#GPGPU
#OTC
#FFT
#3D
#image
#processing
Professor Piotr Targowski and Marcin Sylwestrzak
Optical coherence tomography (OCT) is a non-invasive, non-contact technique of optical sectioning of partially transparent objects with micron axial resolution. This relatively new method for depth-resolved non-invasive imaging of weakly scattering objects originates from medicine and has been successfully used as a diagnostic tool in ophthalmology for about a decade. Over this time a significant progress has been made in achieved resolution, sensitivity, data collection speed and post-processing algorithms. As using near-infrared light to unravel internal structure of examined objects, this technique permits localization of all discontinuities and other rapid changes of refractive index within the medium. The result of examination is usually presented in a convenient and easy to analyze manner of cross-sectional view. By serially collecting many such images, volume information may be extracted. Typical OCT instrument comprises a broad-band light source or very fast swept laser, a Michelson interferometer with an investigated object in one arm and the reference mirror in the other. In a case of most developed Fourier domain OCT the information of the internal structure of the object under investigation is encoded in the frequency of interference fringes superimposed over the light source spectrum. The spectral signal is registered by means of a spectrograph (Spectral OCT), or if a very fast-tuned laser is used as a source, with a single photodiode detector (Swept Source OCT). In both modalities, the spectrum of the interference signal is analyzed by means of Fourier transformation, and the components obtained have frequencies proportional to positions of scattering centers within the object.
A short introduction to the technique was given, the apparatus used by the authors was discussed briefly, and a paradigm for reading OCT tomograms was provided. The lecture focused on signal processing in Fourier domain OCT. Firstly the main sequence of procedures leading from the registered spectrum to one line of the tomogram (the A-scan) was described. Apart of Fourier transformation such procedures as linearization to optical frequency scale, spectrum shaping and numerical dispersion compensation are necessary. On this background, the application of massively parallel processing of OCT data with aid of the low-cost Graphic Processing Unit (GPU) was presented. The reported system may be used for real-time imaging of high resolution 2D tomograms or for presentation of volume data. The description of data flow and parallel processing organization in GPU was given. Then some advanced techniques of visualisation of 3D OCT data with use of the OpenGL platform (Open Graphics Library), an Application Programming Interface (API) for writing software dedicated to interactive 3D computer graphics, was presented. Specifically, the utilisation of texture rendering with a modulation of transparency and colour coding as a function of elevations within the object was shown. Finally, more advanced techniques used in functional OCT for revealing flow velocities in blood vessels of human eye were presented.
Add Event to Calendar
