Three-dimensional Reconstruction Of The Frequency Domain Optical Coherence Tomography Images

Compared with the time-domain optical coherence tomography (OCT), frequency domain optical coherence tomography (FDOCT) has higher imaging speed and sensitivity because of its parallel detection and no need of the axial scanning. Hence FDOCT is the main method used in the field of OCT. However, the high imaging speed generates huge data and the real-time recording, processing and three-dimensional display have become key problems for practical applications of FDOCT investigations.In this thesis, both the development of FDOCT and three-dimensional image reconstruction were discussed in the first place. Then the principles of FDOCT and volume visualization were described. Two-dimensional cross-sectional images obtained in our lab were preprocessed, including filtering, suppression of the noise, and interpolation between slices layer. The MATLAB was used to realize the volume visualization. Finally, the ways of utilizing the GPU were discussed to accelerate the volume reconstruction speed. The methods of improving the imaging speed were proposed:when the constraint is PCI bandwidth, the data transfers are parallelized from the host to device with processing kernels to hide memory transfer latency; when the constraint is processing speed, a dual-GPUS architecture can be used to assign different tasks to each GPU and make the system more stable.