Fluorescence microscopy has been a valuable tool in the field of biological science as it allows one to study the structure and interaction of protein complexes and organelles in living cells. However, conventional optical microscopy technique has been limited by a trade-off between spatiotemporal resolution, signal contrast, and photodamage to the biological samples. It means that an increase in spatial resolution or signal contrast comes at the cost of higher laser power, serial-scanning, or longer image acquisition time. Unfortunately, this leads to severe photobleaching and photodamage to the samples and/or limited throughput of imaging, which is highly challenging to be circumvented through only optical imaging technique. Therefore, one has turned to artificial intelligence (AI) in image processing, applying deep learning algorithms to different imaging modalities to overcome these traditional limitations in optical microscopy systems. Herein we present multiple strategies on how deep learning can be applied to solve challenging and fundamental problems in different fluorescence microscopy modalities. To do so, we present UNet-RCAN, a two-step deep learning network architecture based on a residual U-Net and residual channel attention network (RCAN) for image restoration. We demonstrate that UNet-RCAN achieves higher prediction accuracy compared to other state-of-the-art deep learning algorithms while maintaining the resolution of an output image compared to ground-truth data acquired with optical microscopes. We applied our method to three fluorescence imaging modalities. Firstly, we successfully demonstrate that UNet-RCAN can achieve up to two orders of magnitude acceleration in stimulated emission depletion (STED) imaging while maintaining super-resolution. This significant acceleration enables mitigation of photobleaching and photodamage by robust restoration of noisy 2D and 3D STED images from multiple targets as well as live-cell STED imaging of inner-mitochondrial dynamics with a ten-fold increase in the number of acquired frames compared to conventional STED microscopy. Secondly, we apply our approach in restoring high-resolution widefield deconvolution images of living cells with low light intensity and low photodamage. We show that the accuracy of deconvolution can significantly improve after image restoration with deep learning. Lastly, we show the application of UNet-RCAN in the resolution enhancement of single-shot volumetric imaging with a low numerical aperture objective lens.

Laser scanning confocal microscopy (LSCM) is an optical imaging technique that provides improved resolution and sensitivity over conventional methods of optical microscopy. However, the cost of most commercial LSCM systems exceeds the financial limitations of many smaller laboratories. The design of a custom LSCM created at a fraction of the cost of a commercial model is discussed in this paper. The increase in the incidence rate of skin cancer in the world today is alarming, as such, it is essential to provide an early, rapid and effective method for in vivo diagnostics of human skin tissue. LSCM is capable of detecting alterations in skin morphology and configuration, as well as providing chemical composition information which may be indicative of the development of skin cancer. If developed successfully, LSCM could replace the current invasive biopsy procedures performed today with a quick, non-invasive optical scanning method that would prove beneficial for both patients and physicians alike.

Traditional brain imaging modalities, for example, MRI, CT scan, X-ray, etc. can provide precise and high-resolution images of the brain for diagnosing lesions, tumors or clots inside the brain. However, these modalities require bulky and expensive test setups accessible only at specialized diagnostic centers, and hence may not be suitable or affordable to many patients. Furthermore, the inherent health risks limit the usability of these modalities for frequent monitoring. Microwave imaging is deemed a promising alternative due to its being cost-effective, portable, non-ionizing, non-intrusive. Therefore, this work aims to design an effective microwave head imaging system for the detection of blood clots inside the brain. Two miniaturized antipodal Vivaldi antenna designs are proposed which can provide wideband operation covering the low microwave frequency range (within 1 - 6 GHz) while having electrically small dimensions, directional radiation pattern with reasonable gain, and without requiring immersion in any matching/ coupling liquid. A head imaging system is presented which utilizes a quarter-head scanning approach, to reconstruct four images of the brain by scanning four quarters of the head, using the designed antipodal wideband Vivaldi antenna. A numerical brain model, with and without the presence of blood clot, is simulated using the proposed head-imaging system. At each quarter, the antenna is placed at nine different positions for scanning. The reflected signal at each position is processed and using confocal microwave imaging technique four images of the brain are reconstructed. A comparison is made among the four images in terms of their intensities, for the detection and approximate location of the blood clot inside the brain. The presence of higher intensity regions in any specific quarter of the head demonstrates the presence of a clot and its location and validates the feasibility of the proposed head imaging system using the low frequency wideband Vivaldi antenna.

<p>Mechanical oscillating drills and saws are used in orthopaedic surgery to cut bone and develop screw-holes; however, their use causes friction resulting in significant thermal damage. Ultrashort pulsed lasers appear well-suited to replace traditional tools as they have the ability to efficiently remove bone tissue while causing only minimal collateral damage. Laser ablation also has the added advantages of: (i) no mechanical vibration; (ii) minimal invasiveness; and (iii) small focus spot size. In this thesis work, we experimentally investigated a few key aspects of ultrashort laser ablation of bone tissue.</p> <p>The ablation threshold of unaltered bone was measured using the <em>D</em>²technique and found to range from 1.66 J/cm²± 0.87 J/cm² to 2.37 J/cm²± 0.78 J/cm² depending on incident pulse number. The reduction in ablation threshold with pulse number was an indication of an incubation effect. Using a power law model, the incubation coefficient, ζ, was measured to be 0.89 ± 0.03.</p> <p>The effect of specific laser parameters and drilling protocols on ablation efficiency was also characterized. For ultrashort pulses (≤10 ps), the removal rate was found to be inversely related to the pulse duration; however, irradiation with 5-10 ps pulses were also shown to result in significant tissue removal. With a pulse repetition rate of 1 kHz, the removal rate was observed to be highest when ablating with 50-100 pulses per spot.</p> <p>Larger volumes (>1 mm³) of bone tissue were removed using laser scanning procedures. A series of scanned concentric circles produced a structure ~2.4 mm deep; however, ablated side-lobes were present at oblique angles to the incident beam. A two-layer structure subsequently produced no side-lobes. The ablative precision in trabecular bone was observed to be less than cortical bone. Using mimicked Nd:YAG laser parameters, cylindrical drilling produced craters significantly less deep than those achieved with a typical Ti:Sapphire configuration. The ability to drill large-scale holes using low average pulse energies and optimized scanning procedures will alleviate the stringent requirements for optical components in clinical practice.</p> / Master of Science (MSc)

<p>Current gold standard practices for the diagnosis of tissue disease involve invasive tissue biopsies subjected to a time consuming histopathological examination process. An optical biopsy can offer a non-invasive diagnostic alternative by exploiting the properties of naturally occurring light-tissue interactions. A time-resolved fluorescence spectroscopy instrument (355 nm excitation) has previously been developed by our lab to capture the fluorescence response of gastrointestinal tissue (370-550 nm in 5 nm increments, 25 ns at 1000 ps/pt). Measurements were conducted ex-vivo during routine upper gastrointestinal tract biopsies on duodenum, antrum, stomach body, and esophageal tissue. The work currently presented is focused on protocol development for tissue handling, measurement collection, clinical data management, fluorescent decay modeling using Laguerre based deconvolution, instrument performance evaluation, and k-means based classification.</p> <p>Descriptive parameters derived from spectral (total signal intensity) and temporal (lifetime and Laguerre polynomial coefficients) analysis were used to evaluate the data. It was found that data were only compromised when the total signal intensity for the peak wavelength 455 nm fell blow 19.5 V·ns. The data did not exhibit any signs of photobleaching or pulse width broadening that would have otherwise distorted the lifetime from its true fluorescence response. Data for diseased tissue were limited so the clinical diagnosis was used to classify normal duodenum tissue from normal esophageal tissue. Over 400 pairs of parameters demonstrated k-means can identify duodenum tissue with 87.5 % sensitivity and 87.5 % specificity or better. With some dimensional axis transformations these results could be improved. The lifetimes are not factors here but the relative intensity and decay shape were. Protocols can be applied to diseased or other tissue types with little adaptation. Just a single set of parameters may hold the key to help surgeons choose optimum locations for traditional biopsies or perhaps one day replace them altogether.</p> / Master of Applied Science (MASc)

<p>Quantitatively measuring oxygen saturation is important to characterize the physiological or pathological state of tissue function. In this thesis, we demonstrate the possibility of using susceptibility mapping to noninvasively estimate the venous blood oxygen saturation level. Accurate susceptibility quantification is the key to oxygen saturation quantification. Two approaches are presented in this thesis to generate accurate and artifact free susceptibility maps (SM): a regularized inverse filter and a k-space iterative method. Using the regularized inverse filter, with sufficient resolution, major veins in the brain can be visualized. We found that different sized vessels show a different level of contrast depending on their partial volume effects; larger vessels show a bias toward a reduced susceptibility approaching 90% of the expected value. Also, streaking artifacts associated with high susceptibility structures such as veins are obvious in the reconstructed SM. To further improve susceptibility quantification and reduce the streaking artifacts in the SMs, we proposed a threshold-based k-space iterative approach that used geometric information from the SM itself as a constraint to overcome the ill-posed nature of the inverse filter. Both simulations and in vivo results show that most streaking artifacts inside the SM were suppressed by the iterative approach. In simulated data, the bias toward lower mean susceptibility values inside vessels has been shown to decrease from around 10% to 2% when choosing an appropriate threshold value for the proposed iterative method, which brings us one step closer to a practical means to map out oxygen saturation in the brain.</p> / Doctor of Philosophy (PhD)

<p>The annual incidence rate of tumours in the brain and central nervous system (CNS) was 19.89 per 100,000 persons between 2004 and 2008 in the United States. Surgery is a common treatment option for brain and CNS tumours. Typically, biopsy followed by histological analysis is used to confirm tumour types and margin during neurosurgery as an intraoperative diagnostic tool. However, this biopsy method is invasive, sampling number limited and not in real-time. To overcome these problems, many minimally invasive optical techniques, called optical biopsies, have been developed towards intraoperative diagnosis.</p> <p>The research work carried out in this dissertation focuses on combining the time-resolved fluorescence (TRF) and diffuse reflectance (DR) spectroscopy towards intraoperative tumour margin detection in neurosurgery. Combining these two modalities allows us to obtain additional contrast features, thus potentially improving the diagnostic accuracy. To achieve this goal, first, a clinically compatible integrated TRF-DR spectroscopy instrument was developed for <em>in vivo</em> brain tumour study. An acousto-optical-tunable-filter-based spectrometer was designed to acquire the time-resolved fluorescence signal. A dual-modality fibre optic probe was used to collect the TRF and DR signals in a small volume. The system’s capabilities of resolving fluorescence spectrum and lifetime, and optical properties were characterized and validated using tissue phantoms. Second, in order to retrieve the fluorescence impulse response function accurately from measured fluorescence signals, a robust Laguerre-based deconvolution method was optimized by using the constrained linear least squares fitting and high order Laguerre function basis. This optimized Laguerre-based deconvolution method overcomes the over-fitting problem introduced by low signal-to-noise ratio and complex fitting model. Third, an <em>ex vivo</em> clinical study of brain tumours was carried out using the TRF and DR spectroscopy. Fluorescence spectra and lifetime features were selected to classify various tumour types. The sensitivity and specificity of meningioma grade I differentiated from meningioma grade II are both 100%. Finally, in order to increase the measurement tissue volume and obtain imaging contrast features, a scanning-based hyperspectral fluorescence lifetime imaging system was developed. This setup can provide time-, space-, spectrum- resolved multi-dimensional images for tumour margin detection.</p> / Doctor of Philosophy (PhD)

Positron Emission Tomography (PET) is an imaging modality that has become increasingly beneficial in Radiotherapy by improving treatment planning (1). PET reveals tumor volumes that are not well visualized on computed tomography CT or MRI, recognizes metastatic disease, and assesses radiotherapy treatment (1). It also reveals areas of the tumor that are more radiosensitive allowing for dose painting - a non-homogenous dose treatment across the tumor (1). However, PET is not without limitations. The quantitative unit of PET images, the Standardized Uptake Value (SUV), is affected by many factors such as reconstruction algorithm, patient weight, and tracer uptake time (2). In fact, PET is so sensitive that a patient imaged twice in a single day on the same machine and same protocol will produce different SUV values. The objective of this research was to increase the capabilities of PET by exploring other quantitative PET/CT measures for Radiotherapy treatment applications. The technique of quantitative image feature analysis, nowadays known as radiomics, was applied to PET and CT images. Image features were then extracted from PET/CT images and how the features differed between conventional and respiratory-gated PET/CT images in lung cancer was analyzed. The influence of noise on image features was analyzed by applying uncorrelated, Gaussian noise to PET/CT images and measuring how significantly noise affected features. Quantitative PET/CT measures outside of image feature analysis were also investigated. The correlation of esophageal metabolic tumor volumes (tumor volume demonstrating high metabolic uptake) and endoscopically implanted fiducial markers was studied. It was found that certain image features differed greatly between conventional and respiratory-gated PET/CT. The differences were mainly due to the effect of respiratory motion including affine motion, rotational motion and tumor deformation. Also, certain feature groups were more affected by noise than others. For instance, contour-dependent shape features exhibited the least change with noise. Comparatively, GLSZM features exhibited the greatest change with added noise. Discordance was discovered between the inferior and superior tumor fiducial markers and metabolic tumor volume (MTV). This demonstrated a need for both fiducial markers and MTV to provide a comprehensive view of a tumor. These studies called attention to the differences in features caused by factors such as motion, acquisition parameters, and noise, etc. Investigators should be aware of these effects. PET/CT radiomic features are indeed highly affected by noise and motion. For accurate clinical use, these effects must be account by investigators and future clinical users. Further investigation is warranted towards the standardization of PET/CT radiomic feature acquisition and clinical application.

Heart failure is a common condition during which the pumping action of the heart is affected because the heart does not contract or relax properly. Heart failure affects about 5 million Americans, with 550,000 new cases diagnosed each year. Cardiac resynchronization therapy (CRT) is used to treat symptoms and other complications associated with a heart failure. While performing CRT, Implantation of a pacing lead in the left ventricle of the heart is a very challenging surgical procedure performed with fluoroscopy. The target location is often difficult to reach through the tortuous coronary venous anatomy, which varies greatly among individuals. Placement of the pacing lead is an important research topic because the ideal pacing location for some patients with heart disease may be the site of latest contraction in the left ventricle. The purpose of this project is to develop an algorithm to locate and track points on a lead in a sequence of images. The algorithm will track the motion of the points over time and generate displacement plots over time.

<p>In this thesis we consider the need to generate efficient solvers for inverse imaging problems in a way that supports both quality and performance in software, as well as flexibility in the underlying mathematical models. Many problem domains involve large data sizes and rates, and changes in mathematical modelling are limited only by researcher ingenuity and driven by the value of the application. We use a problem in Magnetic Resonance Imaging to illustrate this situation, motivate the need for better software tools and test the tools we develop. The problem is the determination of velocity profiles, think blood-flow patterns, using Phase Contrast Angiography. Despite the name, this method is completely noninvasive, not requiring the injection of contrast agents, but it is too time-consuming with present imaging and computing technology.</p> <p>Our approach is to separate the specification, the mathematical model, from the implementation details required for performance, using a custom language. The Domain Specific Language (DSL) provided to scientists allows for a complete abstraction from the highly optimized generated code. The mathematical DSL is converted to an internal representation we refer to as the Coconut Expression Library. Our expression library uses the directed acyclic graphs as an underlying data structure, which lends itself nicely to our automatic simplifications, differentiation and subexpression elimination. We show how parallelization and other optimizations are encoded as rules which are applied automatically rather than schemes that need to be implemented by the programmer in the low-level implementation. Finally, we present results, both in terms of numerical results and computational performance.</p> / Master of Science (MSc)