Yeung Shun-hin. / Thesis (M.Phil.)--Chinese University of Hong Kong, 2002. / Includes bibliographical references (leaves 52-54). / Abstracts in English and Chinese. / TITLE PAGE --- p.i / THESIS COMMITTEE --- p.ii / ABSTRACT (ENGLISH) --- p.iii / ABSTRACT (CHINESE) --- p.iv / ACKNOWLEDGEMENTS --- p.v / TABLES OF CONTENTS --- p.vi / LIST OF FIGURES --- p.viii / LIST OF TABLES --- p.x / Chapter Chapter 1 --- Introduction --- p.1 / Chapter Chapter 2 --- Computer-controlled Data Acquisition and Frequency Calibration System for a Ti: sapphire laser spectrometer --- p.3 / Chapter Section 2A --- Motivation and Overview --- p.3 / Chapter Section 2B --- The Hardware --- p.5 / Chapter Section 2C --- The Program --- p.12 / Chapter Section 2D --- Summary --- p.27 / Chapter Chapter 3 --- Cavity Enhanced Absorption Spectroscopy Using Phase-Sensitive Detection --- p.28 / Chapter Section 3A --- Motivation --- p.28 / Chapter Section 3B --- Cavity ring-down technique: the background --- p.29 / Chapter Section 3C --- Cavity enhanced absorption spectroscopy: a historical review --- p.34 / Chapter Section 3D --- Experimental Apparatus --- p.37 / Chapter Section 3E --- Results of Performance tests --- p.41 / Chapter Section 3F --- Applications --- p.45 / Chapter Section 3G --- Summary --- p.49 / Chapter Chapter 4 --- Concluding Remarks --- p.50 / REFERENCES --- p.52

Estimating the dense image motion or optical flow on a real-world nonrigid surface is a fundamental research issue in computer vision, and is applicable to a wide range of fields, including medical imaging, computer animation and robotics. However, nonrigid surface tracking is a difficult challenge because complex nonrigid deformation, accompanied by image blur and natural noise, may lead to severe intensity changes to pixels through an image sequence. This violates the basic intensity constancy assumption of most visual tracking methods. In this thesis, we show that local geometric constraints and long term feature matching techniques can improve local motion preservation, and reduce error accumulation in optical flow estimation. We also demonstrate that combining RGB data with additional information from other sensing channels, can improve tracking performance in blurry scenes as well as allow us to create nonrigid ground truth from real world scenes. First, we introduce a local motion constraint based on a laplacian mesh representation of nonrigid surfaces. This additional constraint term encourages local smoothness whilst simultaneously preserving nonrigid deformation. The results show that our method outperforms most global constraint based models on several popular benchmarks. Second, we observe that the inter-frame blur in general video sequences is near linear, and can be roughly represented by 3D camera motion. To recover dense correspondences from a blurred scene, we therefore design a mechanical device to track camera motion and formulate this as a directional constraint into the optical flow framework. This improves optical flow in blurred scenes. Third, inspired by recent developments in long term feature matching, we introduce an optimisation framework for dense long term tracking -- applicable to any existing optical flow method -- using anchor patches. Finally, we observe that traditional nonrigid surface analysis suffers from a lack of suitable ground truth datasets given real-world noise and long image sequences. To address this, we construct a new ground truth by simultaneously capturing both normal RGB and near-infrared images. The latter spectrum contains dense markers, visible only in the infrared, and represents ground truth positions. Our benchmark contains many real-world scenes and properties absent in existing ground truth datasets.

The long-term complications of diabetes can be dramatically reduced with tight glycemic control. Although the current invasive technology for measuring blood glucose is effective, it not well suited for the real-time measurements necessary for tight control. Near infrared (NIR) absorption spectroscopy, coupling with multivariate calibration modeling, can potentially provide portable, rugged and low-cost instrumentation for continuous glucose sensing. An optical microsensor that can be used in conjunction with an ultrafiltration sampling probe is under development for continuous glucose measurements in interstitial fluid (ISF) collected from subcutaneous tissue. The first part of this research focused on the development of an algorithm for eliminating of effect of temperature variance on NIR glucose measurements. Spectra of 80 bovine blood ultrafiltrate samples were collected under five different temperatures by using a Fourier transform (FT) NIR spectrometer. Based on the fundamental properties of digital Fourier filtering, baseline variations created by difference in the temperature of the blood ultrafiltrate samples were shown to be eliminated by using an optimized Gaussian shape filter response function. PLS calibration models combined with digital Fourier filtering provided standard errors of prediction in the range of 0.3-0.4 mM for sample with temperatures between 25-40 °C. Before applying the microsensor to animal or human measurements, a testing platform was designed and constructed for the eventual purpose of evaluating the ability of the microsensor to follow glucose concentration transients. A series of computer-controlled pumps were used in combination with an ultrafiltration probe to create glucose transients and deliver the corresponding samples to the spectrometer for analysis. NIR spectra were collected continuously as the concentrations of glucose, urea, and lactate were varied independently. Glucose transients were followed over periods of days by using either partial least squares (PLS) or net analyte signal (NAS) calibration methods. The NAS calibration method and a modified Hybrid Linear Analysis (HLA) method were investigated for monitoring the concentrations of glucose and lactate during microbial fermentations. An UF-sampling probe is used to collect samples of the fermentation broth and deliver these samples to the spectrometer for continuous analysis. The established NAS and modified-HLA calibration models provided glucose and lactate concentration measurements with mean percentage errors of 2 and 3%, respectively. These calibration functions were demonstrated capable of accurate concentration measurements several days beyond the formal calibration process. Lastly, NIR spectra of whole bovine blood samples were used to demonstrate the ability to measure glucose in blood with different levels of hematocrit. Calibration functions were based on PLS modeling and the effective models were developed for measurements from absorbance and single-beam NIR spectra. The method of multiplicative scatter correction was found to be particularly effective in reducing the impact of light scattering caused by the red blood cells at different hematocrit levels. These findings imply that nondestructive NIR spectroscopy has the potential to measure glucose without consuming blood, thereby reducing phlebotomy blood loss in neonates and potentially decreasing the frequency of red blood cell transfusion for this fragile patient population.

Noninvasive glucose sensing has been studied widely. Near infrared (NIR) absorption spectroscopy and Raman scattering spectroscopy are proposed individually and combined as methods for glucose measurement in a three component sample matrix. In both techniques, the light transmits through human skin and a spectrum is collected. The research described in this thesis is like this. The use of individual NIR spectra data and individual Raman spectra data can give a good prediction ability of the partial least-squares (PLS) calibration model. Since the NIR and Raman spectroscopies have complementary nature of molecular vibrations, the research tried to prove the prediction ability of the PLS calibration model can be improved by combining NIR and Raman spectra data. Two approaches are investigated to ascertain the benefits of combining these spectral methods. First, NIR and Raman spectral data collected from a set of 60 samples concated and used to compute multivariate models based on PLS and net analyte signal (NAS) methods. The performance of models based on concated NIR-Raman spectra are compared to conventional models based on only NIR and only Raman spectra. The second strategy reported in this chapter is the simulated NIR and Raman spectra and computing PLS and NAS models by concating these simulated spectra. Spectral simulation permits systematic variations in noise levels. In both cases, various preprocessing methods are explored to find a suitable way to combine the different spectral types. The result from the real spectra data is that adding low signal-to-noise ratio (SNR) to high SNR spectra would make the calibration models worse. The result from the simulated spectra data is that with the same SNR and the same magnitude of the two spectra, the prediction ability of the calibration model can be improved.

The ability to measure glucose transcutaneously and noninvasively is an exciting prospect. Such a procedure will offer a painless way of glucose self-monitoring improving the lives of people with diabetes by lowering the barriers to optimal glycemic control. The noninvasive measurements involve collecting near-infrared spectra (4000–5000 cm-1; 2.0–2.5 µm) of skin with two optical fibers in a transmission geometry. Previous results indicate that repositioning of the fiber optic interface adversely affects both precision and accuracy of such measurements. Slight movements of the interface increase prediction errors more than 2.5–fold when performed with a stationary rat model. In this dissertation, the chemical heterogeneity of skin tissue is explored as a possible cause for the sensitivity of the measurement to the position of the optical interface. Rat and human skin tissues are mapped by using combination near infrared spectra the to provide distributions of the major components of skin: water, collagen type I protein, fat, keratin protein, and two scattering terms of constant and slope. On the basis of the measured heterogeneity, sets of rat and human skin spectra are simulated to investigate the impact of repositioning the fiber-optic interface. Glucose predictions are analyzed for each location of the interface for a series of partial least squares (PLS) calibration vectors established for different locations on the skin. Significant increases in the measurement errors are observed for the situation where the PLS calibration models are generated from spectra associated with one location of the interface while subsequent measurements are performed at slightly locations of the skin matrix. These increases in prediction errors match the 2.5–fold increase observed in vivo. The impact of replacing the spectrum of bovine fat with spectra of native fat for both rat and human skin samples is established. Principal component analysis (PCA) of the spectral residuals reveals that the magnitude of the spectral residuals and the effects of tissue fat content on the quality of the linear regression were decreased. The key implication of the research detailed in this dissertation is that chemical heterogeneity of skin tissue must be considered in multivariate models intended for noninvasive glucose measurements.

Since the advent of precise semiconductor engineering techniques in the 1960s, considerable effort has been devoted both in academia and private industry to the fabrication and testing of complex structures. In addition to other techniques, molecular beam epitaxy (MBE) has made it possible to create devices with single mono-layer accuracy. This facilitates the design of precise band structures and the selection of specific spectroscopic properties for light source materials. The applications of such engineered structures have made solid state devices common commercial quantities. These applications include solid state lasers, light emitting diodes and light sensors. Band gap engineering has been used to design emitters for many wavelength bands, including the short wavelength (SWIR) infrared region which ranges from 1.5 to 2.5 μm [1]. Practical devices include sensors operating in the 2-2.5 μm range. When designing such a device, necessary concerns include the required bias voltage, operating current, input impedance and especially for emitters, the wall-plug efficiency. Three types of engineered structures are considered in this thesis. These include GaInAsSb quaternary alloy bulk active regions, GaInAsSb multiple quantum well devices (MQW) and GaInAsSb cascaded light emitting diodes. The three structures are evaluated according to specific standards applied to emitters of infrared light. The spectral profiles are obtained with photo or electro-luminescence, for the purpose of locating the peak emission wavelength. The peak wavelength for these specimens is in the 2.2-2.5μm window. The emission efficiency is determined by employing three empirical techniques: current/voltage (IV), radiance/current (LI), and carrier lifetime measurements. The first verifies that the structure has the correct electrical properties, by measuring among other parameters the activation voltage. The second is used to determine the energy efficiency of the device, including the wall-plug and quantum efficiencies. The last provides estimates of the relative magnitude of the Shockley Read Hall, radiative and Auger coefficients. These constants illustrate the overall radiative efficiency of the material, by noting comparisons between radiative and non-radiative recombination rates.

Near IR (NIR) sensitive Au-Au₂S nanoparticles were prepared by mixing HAuCl₄ and Na₂S in aqueous solutions. An anti-tumor drug, cis-platin, was adsorbed onto Au-Au₂S nanoparticle surface via the 11-mercaptoundecanoic acid layers. The results showed that the degree of adsorption of cis-platin onto Au-Au₂S nanoparticles was controlled by the pH value of solution, and the drug release was sensitive to NIR irradiation. The cis-platin loaded Au-Au₂S nanoparticles can be potentially applied as NIR activated drug delivery carrier. / Singapore-MIT Alliance (SMA)

Brain-computer interface (BCI) technologies allow users to control external devices through brain activity alone, circumventing the somatic nervous system and the need for overt physical movement. BCIs may potentially benefit individuals with severe neuromuscular disorders who experience significant, and often total, loss of voluntary muscle control (e.g. amyotrophic lateral sclerosis, multiple sclerosis, brainstem stroke). Though a majority of BCI research to date has focused on electroencephalography (EEG) for brain signal acquisition, recently researchers have noted the potential of an optical imaging technology called near-infrared spectroscopy (NIRS) for BCI applications. This thesis investigates the feasibility of a practical, online optical BCI based on conscious modulation of prefrontal cortex activity through the performance of different cognitive tasks, specifically mental arithmetic (MA) and mental singing (MS). The thesis comprises five studies, each representing a step toward the realization of a practical optical BCI. The first study demonstrates the feasibility of a two-choice synchronized optical BCI based on intentional control states corresponding to MA and MS. The second study explores a more user-friendly alternative - a two-choice system-paced BCI supporting a single intentional control state (either MA or MS) and a natural baseline, or "no-control (NC)", state. The third study investigates the feasibility of a three-choice system-paced BCI supporting both MA and MS, as well as the NC state. The fourth study examines the consistency with which the relevant mental states can be differentiated over multiple sessions. The first four studies involve healthy adult participants; in the final study, the feasibility of optical BCI use by a user with Duchenne muscular dystrophy is explored. In the first study, MA and MS were classified with an average accuracy of 77.2% (n=10), while in the second, MA and MS were differentiated individually from the NC state with average accuracies of 71.2% and 62.7%, respectively (n=7). In the third study, an average accuracy of 62.5% was obtained for the MA vs. MS vs. NC problem (n=4). The fourth study demonstrated that the ability to classify mental states (specifically MA vs. NC) remains consistent across multiple sessions (p=0.67), but that there is intersession variability in the spatiotemporal characteristics that best discriminate the states. In the final study, a two-session average accuracy of 71.1% was achieved in the MA vs. NC classification problem for the participant with Duchenne muscular dystrophy.

Individuals with limited motor control are often unable to interact with their environment. Recently, near-infrared spectroscopy (NIRS) systems have been investigated as potential brain-computer interfaces (BCI). Previous studies examined data offline, preventing users from understanding how their thoughts triggered the NIRS system. This thesis focused on understanding the short-term effects of feedback on user’s ability to learn how to control BCIs. Data were collected from control and experimental groups over seven sessions, as they performed fast singing imagery or mental arithmetic. Significant differences were observed between the control group’s results in non-feedback sessions and the experimental group’s results in feedback sessions. Qualitative results from 3 of the 10 participants suggested they had control of the feedback system. They performed the task with online accuracies of 61% - 88% in the final 2 sessions with feedback. These results suggest that continued investigation of NIRS feedback systems is warranted.

Near-infrared spectroscopy (NIRS) is an emerging non-invasive brain-computer interface (BCI) modality that measures changes in hemoglobin concentrations in neurocortical tissue. Previous NIRS studies have not employed real-time feedback with online classification, a combination which would allow users to alter their mental strategy on the fly. This thesis reports the results of two online studies. The first study contrasted online classification of prefrontal hemodynamics using an artificial neural network (ANN) and a hidden Markov model-based (HMM) classifier. The second study measured the accuracy of an online linear discriminant classifier. In study 1, only the ANN classifier facilitated online classification rates greater than chance (p=0.0289). In study 2, a new feedback system and experimental protocol led to improved classification rates over those of the first study (p=5.1*10^(-5)). While control over instantaneously generated feedback in online NIRS-BCIs has been demonstrated, factors such as user frustration, mental fatigue, and restrictions on ambient lighting may compromise performance.