In this dissertation, we investigated techniques to develop high-performance, multi-functional, optical coherence tomography (OCT) imaging systems. OCT is a non-invasive high-resolution imaging modality that is frequently used in the fields of dermatology, ophthalmology, and otolaryngology. Our interest was to extend its applications by introducing heterogeneous computation systems, and novel signal and image processing techniques. First, we used the technique of applying general-purpose computations on a graphics processing unit (GPGPU) to develop a real-time imaging OCT device. Second, we demonstrated a heterogeneous computation system running on a mobile chip, to develop a portable OCT device. Third, we introduced a novel image filtering method to improve the OCT image quality by removing speckle noise from the OCT image. Finally, we introduced a novel fingerprint scanner, which is capable of robust imaging of fingerprints using a real-time imaging OCT device.
The real-time OCT imaging system was developed by adopting the compute unified device architecture (CUDA) installed graphics processing units (GPUs). Producing OCT images from raw OCT data requires compute-intensive signal processing and image processing tasks, which limits the imaging speed of the OCT system. We used the GPU for the compute-intensive tasks, such as DC subtraction, k-space resampling, spectrum reshaping, Fourier transformation, image enhancement, etc. to process the OCT signal data and obtain clear OCT images in real time. With the implementation of the GPU computing system, our system was capable of producing 370 images per second from raw OCT data, when the size of the images were 512 × 512 pixels. Moreover, we introduced various methods to optimize the GPU processing system. With the application of the optimization methods, further acceleration of the system was allowed, and our system was capable of producing 1,486 OCT images per second. As a result, we were able to develop a real-time 2D and 3D OCT imaging device, and a real-time speckle variance OCT (SV-OCT) device.
For the purpose of developing a portable OCT device, we introduced a mobile chip OCT imaging system. The mobile chip is a processing device that is frequently used with mobile electronics because of its small size and low power consumption. However, its computing performance is physically limited for operating the OCT system in real time. To overcome this limitation and maximize the computational performance of the mobile chip, we introduced a heterogeneous computing system, which uses multiple processing units on the mobile chip. The heterogeneous computing system was developed with the open computing library (OpenCL), and open graphics library (OpenGL), to process OCT data in real time, and develop a portable OCT imaging device. Without our proposed system, it took more than 15 s to produce an OCT image from raw OCT data. However, our proposed system was capable of producing more than 50 OCT images per second from raw OCT data. It was 617 times faster as compared to the system without our proposed scheme.
A novel speckle noise reduction method was introduced to enhance the visibility of morphological features of the sample tissue from OCT images that are corrupted by speckle noise. A nonlocal means (NLM) filter is one state-of-the-art denoising filter. It exploits the presence of similar features in an image and averages those similar features to remove noise. However, a conventional NLM filter shows somewhat inferior noise reduction performance around the edges, suffering from low efficiency of collecting similar features to be averaged. In order to overcome this phenomenon, we proposed an NLM filter with anisotropic nonlinear windows. The proposed filter was evaluated by comparing it to various other denoising filters, such as a conventional NLM filter, median filter, bilateral filter, and Wiener filter. The denoising performances of the different filters were evaluated in terms of the contrast-to-noise ratio (CNR), equivalent number of looks (ENL), speckle suppression index (SSI), and peak signal-to-noise ratio (PSNR). The evaluation indicated that our proposed NLM filter provides superior denoising performance, among the different filters.
A novel fingerprint scanning device was also studied by adopting our real-time OCT device. Our proposed fingerprint scanning device is resistant to an “attack” by fake fingerprints and robustly captures a clear fingerprint image from the fingertips even under poor conditions. It accomplishes this by capturing subsurface layer fingerprints instead of surface layer fingerprints. In order to obtain internal fingerprint images from raw OCT data in real time, we used the GPU for massive parallel computation, along with an automated method for extracting the internal fingerprint from a 3D scan of a fingertip. Our novel spectral-domain OCT (SD-OCT)-based 3D fingerprint scanner system is capable of obtaining an internal fingerprint image within 2 s. Additionally, the robustness of the OCT fingerprint scanner was established by comparing fingerprint images—of wet, stained, and damaged fingertips—that were obtained by the OCT system with those from a commercially available optical fingerprint scanner.
