GPU-accelerated non-uniform fast Fourier transform-based compressive sensing spectral domain optical coherence tomography

Physics-guided deep learning-based real-time image reconstruction of Fourier-domain optical coherence tomography

In this paper, we introduce a physics-guided deep learning approach for high-quality, real-time Fourier-domain optical coherence tomography (FD-OCT) image reconstruction. Unlike traditional supervised deep learning methods, the proposed method employs unsupervised learning. It leverages the underlying OCT imaging physics to guide the neural networks, which could thus generate high-quality images and provide a physically sound solution to the original problem. Evaluations on synthetic and experimental datasets demonstrate the superior performance of our proposed physics-guided deep learning approach. The method achieves the highest image quality metrics compared to the inverse discrete Fourier transform (IDFT), the optimization-based methods, and several state-of-the-art methods based on deep learning. Our method enables real-time frame rates of 232 fps for synthetic images and 87 fps for experimental images, which represents significant improvements over existing techniques. Our physics-guided deep learning-based approach could offer a promising solution for FD-OCT image reconstruction, which potentially paves the way for leveraging the power of deep learning in real-world OCT imaging applications.

Predictive coding compressive sensing optical coherence tomography hardware implementation

Compressed sensing (CS) is an approach that enables comprehensive imaging by reducing both imaging time and data density, and is a theory that enables undersampling far below the Nyquist sampling rate and guarantees high-accuracy image recovery. Prior efforts in the literature have focused on demonstrations of synthetic undersampling and reconstructions enabled by compressed sensing. In this paper, we demonstrate the first physical, hardware-based sub-Nyquist sampling with a galvanometer-based OCT system with subsequent reconstruction enabled by compressed sensing. Acquired images of a variety of samples, with volume scanning time reduced by 89% (12.5% compression rate), were successfully reconstructed with relative error (RE) of less than 20% and mean square error (MSE) of around 1%.

GPU-accelerated iterative method for FD-OCT image reconstruction with an image-level cross-domain regularizer

The image reconstruction for Fourier-domain optical coherence tomography (FD-OCT) could be achieved by iterative methods, which offer a more accurate estimation than the traditional inverse discrete Fourier transform (IDFT) reconstruction. However, the existing iterative methods are mostly A-line-based and are developed on CPU, which causes slow reconstruction. Besides, A-line-based reconstruction makes the iterative methods incompatible with most existing image-level image processing techniques. In this paper, we proposed an iterative method that enables B-scan-based OCT image reconstruction, which has three major advantages: (1) Large-scale parallelism of the OCT dataset is achieved by using GPU acceleration. (2) A novel image-level cross-domain regularizer was developed, such that the image processing could be performed simultaneously during the image reconstruction; an enhanced image could be directly generated from the OCT interferogram. (3) The scalability of the proposed method was demonstrated for 3D OCT image reconstruction. Compared with the state-of-the-art (SOTA) iterative approaches, the proposed method achieves higher image quality with reduced computational time by orders of magnitude. To further show the image enhancement ability, a comparison was conducted between the proposed method and the conventional workflow, in which an IDFT reconstructed OCT image is later processed by a total variation-regularized denoising algorithm. The proposed method can achieve a better performance evaluated by metrics such as signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR), while the speed is improved by more than 30 times. Real-time image reconstruction at more than 20 B-scans per second was realized with a frame size of 4096 (axial) × 1000 (lateral), which showcases the great potential of the proposed method in real-world applications.

3-D compressed sensing optical coherence tomography using predictive coding

We present a compressed sensing (CS) algorithm and sampling strategy for reconstructing 3-D Optical Coherence Tomography (OCT) image volumes from as little as 10% of the original data. Reconstruction using the proposed method, Denoising Predictive Coding (DN-PC), is demonstrated for five clinically relevant tissue types including human heart, retina, uterus, breast, and bovine ligament. DN-PC reconstructs the difference between adjacent b-scans in a volume and iteratively applies Gaussian filtering to improve image sparsity. An a-line sampling strategy was developed that can be easily implemented in existing Spectral-Domain OCT systems and reduce scan time by up to 90%.

Pseudo-real-time retinal layer segmentation for high-resolution adaptive optics optical coherence tomography

We present a pseudo-real-time retinal layer segmentation for high-resolution Sensorless Adaptive Optics-Optical Coherence Tomography (SAO-OCT). Our pseudo-real-time segmentation method is based on Dijkstra's algorithm that uses the intensity of pixels and the vertical gradient of the image to find the minimum cost in a geometric graph formulation within a limited search region. It segments six retinal layer boundaries in an iterative process according to their order of prominence. The segmentation time is strongly correlated to the number of retinal layers to be segmented. Our program permits en face images to be extracted during data acquisition to guide the depth specific focus control and depth dependent aberration correction for high-resolution SAO-OCT systems. The average processing times for our entire pipeline for segmenting six layers in a retinal B-scan of 496 × 400 and 240 × 400 pixels are around 25.60 and 13.76 ms, respectively. When reducing the number of layers segmented to only two layers, the time required for a 240 × 400 pixel image is 8.26 ms.

Structural and Functional Sensing of Bio-Tissues Based on Compressive Sensing Spectral Domain Optical Coherence Tomography

In this paper, a full depth 2D CS-SDOCT approach is proposed, which combines two-dimensional (2D) compressive sensing spectral-domain optical coherence tomography (CS-SDOCT) and dispersion encoding (ED) technologies, and its applications in structural imaging and functional sensing of bio-tissues are studied. Specifically, by introducing a large dispersion mismatch between the reference arm and sample arm in SD-OCT system, the reconstruction of the under-sampled A-scan data and the removal of the conjugated images can be achieved simultaneously by only two iterations. The under-sampled B-scan data is then reconstructed using the classic CS reconstruction algorithm. For a 5 mm × 3.2 mm fish-eye image, the conjugated image was reduced by 31.4 dB using 50% × 50% sampled data (250 depth scans and 480 spectral sampling points per depth scan), and all A-scan data was reconstructed in only 1.2 s. In addition, we analyze the application performance of the CS-SDOCT in functional sensing of locally homogeneous tissue. Simulation and experimental results show that this method can correctly reconstruct the extinction coefficient spectrum under reasonable iteration times. When 8 iterations were used to reconstruct the A-scan data in the imaging experiment of fisheye, the extinction coefficient spectrum calculated using 50% × 50% data was approximately consistent with that obtained with 100% data.

Dual-Channel Spectral Domain Optical Coherence Tomography Based on a Single Spectrometer Using Compressive Sensing

Dual-channel spectral domain optical coherence tomography (SD-OCT) is one of the effective methods for improving imaging depth and imaging speed. In this paper, we design a dual-channel SD-OCT system based on a single spectrometer that can operate in two modes: (1) Increasing imaging speed and (2) expanding imaging depth. An optical path offset is preintroduced between the two channels to separate the two-channel data. However, this offset increases the requirement for the spectral resolution of the spectrometer in mode (1), so compressive sensing (CS) technology is used herein to overcome this problem. Consequently, in mode (1), when the spectral resolution of the spectrometer is the same as that used in the single-channel system, we use a dual-channel SD-OCT system combined with CS technology to double the imaging speed. In mode (2), when the spectral resolution of the spectrometer is only half of that used in a single-channel system, the imaging depth can be nearly doubled. We demonstrate the feasibility and effectiveness of the method proposed in this work by imaging a mirror, a fish fin, a fish eye, and an onion.

Accelerated near-field algorithm of sparse apertures by non-uniform fast Fourier transform

We present an accelerated algorithm for calculating the near-field of non-uniform sparse apertures with non-uniform fast Fourier transform (NUFFT). The distances of the adjacent units in non-uniform sparse apertures are unequal and larger than half a wavelength. The near-field of apertures can be calculated by the angular spectrum method and the convolution methods, and according to the different convolution kernels, the convolution methods can be divided as the Fresnel kernel convolution and the Rayleigh-Sommerfeld kernel convolution. The Fresnel kernel is the approximation of the Rayleigh-Sommerfeld kernel in the far regions of the near-field zone. In uniform apertures, the three methods can be accelerated by fast Fourier transform (FFT). However, FFT should be replaced by NUFFT for non-uniform sparse apertures. The simulation results reveal that the Rayleigh-Sommerfeld convolution with NUFFT (RS-NUFFT) can be applied to all aperture sizes, distributions and near-field distances. After investigating the error sources in RS-NUFFT, the techniques (padding zeros for apertures, increasing sampling rate for the convolution kernel) are developed for increasing the calculation accuracy of RS-NUFFT.

Full-depth compressive sensing spectral-domain optical coherence tomography based on a compressive dispersion encoding method

By combining the advantages of compressive sensing optical coherence tomography (OCT) and full-depth OCT in terms of imaging time and imaging depth, we demonstrate how compressive sampling and dispersion encoding can be simultaneously used to reconstruct a full-depth OCT image. Moreover, by considering the image processing speed, we propose a two-step compressive dispersion encoding (TCDE) method, in which a large dispersion imbalance is introduced between the reference arm and the sample arm and two iterations are performed. The first iteration selects the signals with higher intensity and then removes their conjugate items and incoherent aliasing artifacts caused by undersampling based on the least squares method. The second iteration selects the signals with lower intensity. Experimental results show that nearly the same conjugate inhibition ratio can be obtained with 50% sampled data and 100% sampled data using the TCDE method. Full-depth images of a glass slide, an onion, and a live fish eye are obtained from 50% and 100% sampled data using the TCDE method. For a 1.4  mm×3.6  mm fish eye image, the conjugate items are reduced by 33.2 and 31.7 dB using 50% sampled data and 100% sampled data, respectively.

Large-volume, low-cost, high-precision FMCW tomography using stitched DFBs

Optical frequency-modulated continuous-wave (FMCW) reflectometry is a ranging technique that allows for high-resolution distance measurements over long ranges. Similarly, swept-source optical coherence tomography (SS-OCT) provides high-resolution depth imaging over typically shorter distances and higher scan speeds. In this work, we demonstrate a low-cost, low-bandwidth 3D imaging system that provides the high axial resolution imaging capability normally associated with SS-OCT over typical FMCW ranging depths. The imaging system combines 12 distributed feedback laser (DFB) elements from a single butterfly module to provide an axial resolution of 27.1 μm over 6 m of depth and up to 14 cubic meters of volume. Active sweep linearization is used, greatly reducing the signal processing overhead. Various sub-surface, OCT-style tomograms of semi-transparent objects are shown, as well as 3D maps of various objects over depths ranging from sub-millimeter to several meters. Such imaging capability would make long-distance, high-resolution surface interrogation possible in a low-cost, compact package.

Optofluidic time-stretch imaging - an emerging tool for high-throughput imaging flow cytometry

Optical imaging is arguably the most effective tool to visualize living cells with high spatiotemporal resolution and in a nearly noninvasive manner. Driven by this capability, state-of-the-art cellular assay techniques have increasingly been adopting optical imaging for classifying different cell types/stages, and thus dissecting the respective cellular functions. However, it is still a daunting task to image and characterize cell-to-cell variability within an enormous and heterogeneous population - an unmet need in single-cell analysis, which is now widely advocated in modern biology and clinical diagnostics. The challenge stems from the fact that current optical imaging technologies still lack the practical speed and sensitivity for measuring thousands to millions of cells down to the single-cell precision. Adopting the wisdom in high-speed fiber-optics communication, optical time-stretch imaging has emerged as a completely new optical imaging concept which is now proven for ultrahigh-throughput optofluidic single-cell imaging, at least 1-2 orders-of-magnitude higher (up to ∼100 000 cells per second) compared to the existing imaging flow cytometers. It also uniquely enables quantification of intrinsic biophysical markers of individual cells - a largely unexploited class of single-cell signatures that is known to be correlated with the overwhelmingly investigated biochemical markers. With the aim of reaching a wider spectrum of experts specializing in cellular assay developments and applications, this paper highlights the essential basics of optical time-stretch imaging, followed by reviewing the recent developments and applications of optofluidic time-stretch imaging. We will also discuss the current challenges of this technology, in terms of providing new insights in basic biology and enriching the clinical diagnostic toolsets.

Fast compressive measurements acquisition using optimized binary sensing matrices for low-light-level imaging

Compressive measurements benefit low-light-level imaging (L³-imaging) due to the significantly improved measurement signal-to-noise ratio (SNR). However, as with other compressive imaging (CI) systems, compressive L³-imaging is slow. To accelerate the data acquisition, we develop an algorithm to compute the optimal binary sensing matrix that can minimize the image reconstruction error. First, we make use of the measurement SNR and the reconstruction mean square error (MSE) to define the optimal gray-value sensing matrix. Then, we construct an equality-constrained optimization problem to solve for a binary sensing matrix. From several experimental results, we show that the latter delivers a similar reconstruction performance as the former, while having a smaller dynamic range requirement to system sensors.

GPU-accelerated compressive holography

In this paper, we show fast signal reconstruction for compressive holography using a graphics processing unit (GPU). We implemented a fast iterative shrinkage-thresholding algorithm on a GPU to solve the ℓ₁ and total variation (TV) regularized problems that are typically used in compressive holography. Since the algorithm is highly parallel, GPUs can compute it efficiently by data-parallel computing. For better performance, our implementation exploits the structure of the measurement matrix to compute the matrix multiplications. The results show that GPU-based implementation is about 20 times faster than CPU-based implementation.

Iterative l(1)-min algorithm for fixed pattern noise removal in fiber-bundle-based endoscopic imaging

In this study, we developed a signal processing method for fixed pattern noise removal in fiber-bundle-based endoscopic imaging. We physically acquired the fixed pattern of the fiber bundle and used it as a prior image in an l₁ norm minimization (l₁-min) algorithm. We chose an iterative shrinkage thresholding algorithm for l₁ norm minimization. In addition to fixed pattern noise removal, this method also improved image contrast while preserving spatial resolution. The effectiveness of this method was demonstrated on images obtained from a dark-field illuminated reflectance fiber-optic microscope (DRFM). The iterative l₁-min algorithm presented in this paper, in combination with the DRFM system that we previously developed, enables high-resolution, high-sensitivity, intrinsic-contrast, and in situ cellular imaging which has great potential in clinical diagnosis and biomedical research.

Quantitative monitoring of laser-treated engineered skin using optical coherence tomography

Nowadays, laser therapy is a common method for treating various dermatological troubles such as acne and wrinkles because of its efficient and immediate skin enhancement. Although laser treatment has become a routine procedure in medical and cosmetic fields, the prevention of side-effects, such as hyperpigmentation, redness and burning, still remains a critical issue that needs to be addressed. In order to reduce the side-effects while attaining efficient therapeutic outcomes, it is essential to understand the light-skin interaction through evaluation of physiological changes before and after laser therapy. In this study, we introduce a quantitative tissue monitoring method based on optical coherence tomography (OCT) for the evaluation of tissue regeneration after laser irradiation. To create a skin injury model, we applied a fractional CO2 laser on a customized engineered skin model, which is analogous to human skin in terms of its basic biological function and morphology. The irradiated region in the skin was then imaged by a high-speed OCT system, and its morphologic changes were analyzed by automatic segmentation software. Volumetric OCT images in the laser treated area clearly visualized the wound healing progress at different time points and provided comprehensive information which cannot be acquired through conventional monitoring methods. The results showed that the laser wound in engineered skins was mostly recovered from within 1~2 days with a fast recovery time in the vertical direction. However, the entire recovery period varied widely depending on laser doses and skin type. Our results also indicated that OCT-guided laser therapy would be a very promising protocol for optimizing laser treatment for skin therapy.

Tracing the trajectory of photons through Fourier spectrum

By slightly vibrating the mirrors in an interferometer at different frequencies, the photons' trajectory information is stored in the light beam. To read out this information, we record the centroid location of the intensity distribution of output beam and Fourier analyze its time evolution. It is shown that every vibrating mirror contributes a peak in the Fourier spectrum. In other words, we can reveal the trajectory of the photons by figuring out the vibrating mirrors which ever interact with the light beam based on the Fourier spectrum. This techniques is not limited by the vibration amplitude of the mirrors.

Volumetric (3D) compressive sensing spectral domain optical coherence tomography

In this work, we proposed a novel three-dimensional compressive sensing (CS) approach for spectral domain optical coherence tomography (SD OCT) volumetric image acquisition and reconstruction. Instead of taking a spectral volume whose size is the same as that of the volumetric image, our method uses a sub set of the original spectral volume that is under-sampled in all three dimensions, which reduces the amount of spectral measurements to less than 20% of that required by the Shan-non/Nyquist theory. The 3D image is recovered from the under-sampled spectral data dimension-by-dimension using the proposed three-step CS reconstruction strategy. Experimental results show that our method can significantly reduce the sampling rate required for a volumetric SD OCT image while preserving the image quality.

Real-time dispersion-compensated image reconstruction for compressive sensing spectral domain optical coherence tomography

In this work, we propose a novel dispersion compensation method that enables real-time compressive sensing (CS) spectral domain optical coherence tomography (SD OCT) image reconstruction. We show that dispersion compensation can be incorporated into CS SD OCT by multiplying the dispersion-correcting terms by the undersampled spectral data before CS reconstruction. High-quality SD OCT imaging with dispersion compensation was demonstrated at a speed in excess of 70 frames per s using 40% of the spectral measurements required by the well-known Shannon/Nyquist theory. The data processing and image display were performed on a conventional workstation having three graphics processing units.