Advanced Monte Carlo Simulation and Machine Learning for Frequency Domain Optical Coherence Tomography

Optical Coherence Tomography(OCT) is a popular, rapidly growing imaging technique with an increasing number of bio-medical applications due to its noninvasive nature. However, there are three major challenges in understanding and improving an OCT system: (1) Obtaining an OCT image is not easy. It either takes a real medical experiment or requires days of computer simulation. Without much data, it is difficult to study the physical processes underlying OCT imaging of different objects simply because there aren't many imaged objects. (2) Interpretation of an OCT image is also hard. This challenge is more profound than it appears. For instance, it would require a trained expert to tell from an OCT image of human skin whether there is a lesion or not. This is expensive in its own right, but even the expert cannot be sure about the exact size of the lesion or the width of the various skin layers. The take-away message is that analyzing an OCT image even from a high level would usually require a trained expert, and pixel-level interpretation is simply unrealistic. The reason is simple: we have OCT images but not their underlying ground-truth structure, so there is nothing to learn from. (3) The imaging depth of OCT is very limited (millimeter or sub-millimeter on human tissues). While OCT utilizes infrared light for illumination to stay noninvasive, the downside of this is that photons at such long wavelengths can only penetrate a limited depth into the tissue before getting back-scattered. To image a particular region of a tissue, photons first need to reach that region. As a result, OCT signals from deeper regions of the tissue are both weak (since few photons reached there) and distorted (due to multiple scatterings of the contributing photons). This fact alone makes OCT images very hard to interpret.

This thesis addresses the above challenges by successfully developing an advanced Monte Carlo simulation platform which is 10000 times faster than the state-of-the-art simulator in the literature, bringing down the simulation time from 360 hours to a single minute. This powerful simulation tool not only enables us to efficiently generate as many OCT images of objects with arbitrary structure and shape as we want on a common desktop computer, but it also provides us the underlying ground-truth of the simulated images at the same time because we dictate them at the beginning of the simulation. This is one of the key contributions of this thesis. What allows us to build such a powerful simulation tool includes a thorough understanding of the signal formation process, clever implementation of the importance sampling/photon splitting procedure, efficient use of a voxel-based mesh system in determining photon-mesh interception, and a parallel computation of different A-scans that consist a full OCT image, among other programming and mathematical tricks, which will be explained in detail later in the thesis.

Next we aim at the inverse problem: given an OCT image, predict/reconstruct its ground-truth structure on a pixel level. By solving this problem we would be able to interpret an OCT image completely and precisely without the help from a trained expert. It turns out that we can do much better. For simple structures we are able to reconstruct the ground-truth of an OCT image more than 98% correctly, and for more complicated structures (e.g., a multi-layered brain structure) we are looking at 93%. We achieved this through extensive uses of Machine Learning. The success of the Monte Carlo simulation already puts us in a great position by providing us with a great deal of data (effectively unlimited), in the form of (image, truth) pairs. Through a transformation of the high-dimensional response variable, we convert the learning task into a multi-output multi-class classification problem and a multi-output regression problem. We then build a hierarchy architecture of machine learning models (committee of experts) and train different parts of the architecture with specifically designed data sets. In prediction, an unseen OCT image first goes through a classification model to determine its structure (e.g., the number and the types of layers present in the image); then the image is handed to a regression model that is trained specifically for that particular structure to predict the length of the different layers and by doing so reconstruct the ground-truth of the image. We also demonstrate that ideas from Deep Learning can be useful to further improve the performance.

It is worth pointing out that solving the inverse problem automatically improves the imaging depth, since previously the lower half of an OCT image (i.e., greater depth) can be hardly seen but now becomes fully resolved. Interestingly, although OCT signals consisting the lower half of the image are weak, messy, and uninterpretable to human eyes, they still carry enough information which when fed into a well-trained machine learning model spits out precisely the true structure of the object being imaged. This is just another case where Artificial Intelligence (AI) outperforms human. To the best knowledge of the author, this thesis is not only a success but also the first attempt to reconstruct an OCT image at a pixel level. To even give a try on this kind of task, it would require fully annotated OCT images and a lot of them (hundreds or even thousands). This is clearly impossible without a powerful simulation tool like the one developed in this thesis.

Endoscopic optical coherence tomography : design and application

This thesis presents an investigation on endoscopic optical coherence tomography (OCT). As a noninvasive imaging modality, OCT emerges as an increasingly important diagnostic tool for many clinical applications. Despite of many of its merits, such as high resolution and depth resolvability, a major limitation is the relatively shallow penetration depth in tissue (about 2∼3 mm). This is mainly due to tissue scattering and absorption. To overcome this limitation, people have been developing many different endoscopic OCT systems. By utilizing a minimally invasive endoscope, the OCT probing beam can be brought to the close vicinity of the tissue of interest and bypass the scattering of intervening tissues so that it can collect the reflected light signal from desired depth and provide a clear image representing the physiological structure of the region, which can not be disclosed by traditional OCT. In this thesis, three endoscope designs have been studied. While they rely on vastly different principles, they all converge to solve this long-standing problem.

A hand-held endoscope with manual scanning is first explored. When a user is holding a hand- held endoscope to examine samples, the movement of the device provides a natural scanning. We proposed and implemented an optical tracking system to estimate and record the trajectory of the device. By registering the OCT axial scan with the spatial information obtained from the tracking system, one can use this system to simply ‘paint’ a desired volume and get any arbitrary scanning pattern by manually waving the endoscope over the region of interest. The accuracy of the tracking system was measured to be about 10 microns, which is comparable to the lateral resolution of most OCT system. Targeted phantom sample and biological samples were manually scanned and the reconstructed images verified the method.

Next, we investigated a mechanical way to steer the beam in an OCT endoscope, which is termed as Paired-angle-rotation scanning (PARS). This concept was proposed by my colleague and we further developed this technology by enhancing the longevity of the device, reducing the diameter of the probe, and shrinking down the form factor of the hand-piece. Several families of probes have been designed and fabricated with various optical performances. They have been applied to different applications, including the collector channel examination for glaucoma stent implantation, and vitreous remnant detection during live animal vitrectomy.

Lastly a novel non-moving scanning method has been devised. This approach is based on the EO effect of a KTN crystal. With Ohmic contact of the electrodes, the KTN crystal can exhibit a special mode of EO effect, termed as space-charge-controlled electro-optic effect, where the carrier electron will be injected into the material via the Ohmic contact. By applying a high voltage across the material, a linear phase profile can be built under this mode, which in turn deflects the light beam passing through. We constructed a relay telescope to adapt the KTN deflector into a bench top OCT scanning system. One of major technical challenges for this system is the strong chromatic dispersion of KTN crystal within the wavelength band of OCT system. We investigated its impact on the acquired OCT images and proposed a new approach to estimate and compensate the actual dispersion. Comparing with traditional methods, the new method is more computational efficient and accurate. Some biological samples were scanned by this KTN based system. The acquired images justified the feasibility of the usage of this system into a endoscopy setting. My research above all aims to provide solutions to implement an OCT endoscope. As technology evolves from manual, to mechanical, and to electrical approaches, different solutions are presented. Since all have their own advantages and disadvantages, one has to determine the actual requirements and select the best fit for a specific application.

Two roads to planet detection

One of the most exciting discoveries in astrophysics of the last last decade is of the sheer diversity of planetary systems. These include "hot Jupiters", giant planets so close to their host stars that they orbit once every few days; "Super-Earths", planets with sizes intermediate to those of Earth and Neptune, of which no analogs exist in our own solar system; multi-planet systems with planets smaller than Mars to larger than Jupiter; planets orbiting binary stars; free-floating planets flying through the emptiness of space without any star; even planets orbiting pulsars. Despite these remarkable discoveries, the field is still young, and there are many areas about which precious little is known. In particular, we don't know the planets orbiting Sun-like stars nearest to our own solar system, and we know very little about the compositions of extrasolar planets. This thesis provides developments in those directions, through two instrumentation projects.

The first chapter of this thesis concerns detecting planets in the Solar neighborhood using precision stellar radial velocities, also known as the Doppler technique. We present an analysis determining the most efficient way to detect planets considering factors such as spectral type, wavelengths of observation, spectrograph resolution, observing time, and instrumental sensitivity. We show that G and K dwarfs observed at 400-600 nm are the best targets for surveys complete down to a given planet mass and out to a specified orbital period. Overall we find that M dwarfs observed at 700-800 nm are the best targets for habitable-zone planets, particularly when including the effects of systematic noise floors caused by instrumental imperfections. Somewhat surprisingly, we demonstrate that a modestly sized observatory, with a dedicated observing program, is up to the task of discovering such planets.

We present just such an observatory in the second chapter, called the "MINiature Exoplanet Radial Velocity Array," or MINERVA. We describe the design, which uses a novel multi-aperture approach to increase stability and performance through lower system etendue, as well as keeping costs and time to deployment down. We present calculations of the expected planet yield, and data showing the system performance from our testing and development of the system at Caltech's campus. We also present the motivation, design, and performance of a fiber coupling system for the array, critical for efficiently and reliably bringing light from the telescopes to the spectrograph. We finish by presenting the current status of MINERVA, operational at Mt. Hopkins observatory in Arizona.

The second part of this thesis concerns a very different method of planet detection, direct imaging, which involves discovery and characterization of planets by collecting and analyzing their light. Directly analyzing planetary light is the most promising way to study their atmospheres, formation histories, and compositions. Direct imaging is extremely challenging, as it requires a high performance adaptive optics system to unblur the point-spread function of the parent star through the atmosphere, a coronagraph to suppress stellar diffraction, and image post-processing to remove non-common path "speckle" aberrations that can overwhelm any planetary companions.

To this end, we present the "Stellar Double Coronagraph," or SDC, a flexible coronagraphic platform for use with the 200" Hale telescope. It has two focal and pupil planes, allowing for a number of different observing modes, including multiple vortex phase masks in series for improved contrast and inner working angle behind the obscured aperture of the telescope. We present the motivation, design, performance, and data reduction pipeline of the instrument. In the following chapter, we present some early science results, including the first image of a companion to the star delta Andromeda, which had been previously hypothesized but never seen.

A further chapter presents a wavefront control code developed for the instrument, using the technique of "speckle nulling," which can remove optical aberrations from the system using the deformable mirror of the adaptive optics system. This code allows for improved contrast and inner working angles, and was written in a modular style so as to be portable to other high contrast imaging platforms. We present its performance on optical, near-infrared, and thermal infrared instruments on the Palomar and Keck telescopes, showing how it can improve contrasts by a factor of a few in less than ten iterations.

One of the large challenges in direct imaging is sensing and correcting the electric field in the focal plane to remove scattered light that can be much brighter than any planets. In the last chapter, we present a new method of focal-plane wavefront sensing, combining a coronagraph with a simple phase-shifting interferometer. We present its design and implementation on the Stellar Double Coronagraph, demonstrating its ability to create regions of high contrast by measuring and correcting for optical aberrations in the focal plane. Finally, we derive how it is possible to use the same hardware to distinguish companions from speckle errors using the principles of optical coherence. We present results observing the brown dwarf HD 49197b, demonstrating the ability to detect it despite it being buried in the speckle noise floor. We believe this is the first detection of a substellar companion using the coherence properties of light.

High-coherence hybrid Si/III-V semiconductor lasers

The relentlessly increasing demand for network bandwidth, driven primarily by Internet-based services such as mobile computing, cloud storage and video-on-demand, calls for more efficient utilization of the available communication spectrum, as that afforded by the resurging DSP-powered coherent optical communications. Encoding information in the phase of the optical carrier, using multilevel phase modulationformats, and employing coherent detection at the receiver allows for enhanced spectral efficiency and thus enables increased network capacity. The distributed feedback semiconductor laser (DFB) has served as the near exclusive light source powering the fiber optic, long-haul network for over 30 years. The transition to coherent communication systems is pushing the DFB laser to the limits of its abilities. This is due to its limited temporal coherence that directly translates into the number of different phases that can be imparted to a single optical pulse and thus to the data capacity. Temporal coherence, most commonly quantified in the spectral linewidth Δν, is limited by phase noise, result of quantum-mandated spontaneous emission of photons due to random recombination of carriers in the active region of the laser.

In this work we develop a generically new type of semiconductor laser with the requisite coherence properties. We demonstrate electrically driven lasers characterized by a quantum noise-limited spectral linewidth as low as 18 kHz. This narrow linewidth is result of a fundamentally new laser design philosophy that separates the functions of photon generation and storage and is enabled by a hybrid Si/III-V integration platform. Photons generated in the active region of the III-V material are readily stored away in the low loss Si that hosts the bulk of the laser field, thereby enabling high-Q photon storage. The storage of a large number of coherent quanta acts as an optical flywheel, which by its inertia reduces the effect of the spontaneous emission-mandated phase perturbations on the laser field, while the enhanced photon lifetime effectively reduces the emission rate of incoherent quanta into the lasing mode. Narrow linewidths are obtained over a wavelength bandwidth spanning the entire optical communication C-band (1530-1575nm) at only a fraction of the input power required by conventional DFB lasers. The results presented in this thesis hold great promise for the large scale integration of lithographically tuned, high-coherence laser arrays for use in coherent communications, that will enable Tb/s-scale data capacities.