This master thesis investigates different approaches for integrating prior information into a neural network for segmentation of medical images. In the study, liver and liver tumor segmentation is performed in a cascading fashion. Context channels in the form of previous segmentations are integrated into a segmentation network at multiple positions and network depths using different integration strategies. Comparisons are made with the traditional integration approach where an input image is concatenated with context channels at a network’s input layer. The aim is to analyze if context information is lost in the upper network layers when the traditional approach is used, and if better results can be achieved if prior information is propagated to deeper layers. The intention is to support further improvements in interactive image segmentation where extra input channels are common. The results that are achieved are, however, inconclusive. It is not possible to differentiate the methods from each other based on the quantitative results, and all the methods show the ability to generalize to an unseen object class after training. Compared to the other evaluated methods there are no indications that the traditional concatenation approach is underachieving, and it cannot be declared that meaningful context information is lost in the deeper network layers.

<p> The central idea behind digital image processing is quite simple. The digital image is fed into a computer one pixel at a time. The computer is programmed to insert these data into an equation, or series of equations, and then store the results of the computation for each pixel. These results form a new digital image that may be displayed or recorded in pictorial format (specific for the particular image processing system in use) or may itself be further manipulated by additional programs. The possible forms of digital image manipulation are literally infinite. The purpose of this project is to implement some image processing techniques to facilitate the image analysis research conducted in the Institute for Materials Research at McMaster University. </p> / Thesis / Master of Science (MSc)

No description available.

The present study examines the role of turbulence on the incipient motion of sediment. For this purpose, well-controlled experiments are performed at the laboratory in a tilting flume. In these tests glass beads of the same size and density are used as the testing material to isolate the role of turbulence. State of the art equipment are used during the course of this study. Specifically, a 3-D Laser Doppler Velocimetry system is employed to measure the instantaneous velocity components at different points near the vicinity of a ball while the ball motion is monitored with a video camera. An image analysis program is developed here to analyze the motion of the particles within a test area. To examine the importance of the different stress components in the entrainment of sediment, five tests of different packing configuration are performed. Specifically three different roughness regimes are examined namely, the isolated, the wake interference, and the skimming flow. The results reveal that the instantaneous normal stress in the streamwise direction is the most dominant component of the instantaneous stress tensor. The backbone of this study is the development of a methodology to link the effects of turbulence with the commencement of sediment motion. It is considered that the metastable bursting cycle (i.e. sweeps, ejections, inward and outward interactions) is responsible for the sediment entrainment. And that the sediment entrainment, if any, occurs within a bursting period. The main concept behind the determination of the critical conditions is that the probability of the entrainment of sediment (effect) is equal to the probability of occurrence of these highly energetic turbulent events that have magnitude greater than the critical (cause). The probability of sediment entrainment is computed by means of the image analysis tool. The balance of moments is obtained here to determine the minimum moment that is required for the commencement of sediment motion. The balance of moments yields the deduction of a new variable that is used to describe the probability of occurrence of the different turbulent events. This variable is the summation of the instantaneous normal stresses in the streamwise and vertical direction. It is shown here that a two-parameter gamma density function describes quite well the statistical behavior of this variable. The results that are obtained from the existing model suggest that the present methodology can adequately describe the commencement of sediment motion. It is shown here that the traditionally used shear stress term uw may not be the appropriate measure for the determination of the critical conditions. / Ph. D.

The implementation of process control in fine coal processing operations has traditionally been limited by the lack of adequate on-line ash sensors. Several nuclear based analyzers are available, yet none have seen widespread acceptance by the coal industry. This is due largely to their high cost, the influences of seam type and pyrite content on accuracy, and the inconvenience of having radioactive sources in a plant. Thus, reliable process control of fine coal circuits is often unobtainable due to the lack of on-line monitoring devices for ash content in process slurry streams. Recently, a video-based slurry sensor for ash analysis of coal tailings has been developed which provides a low cost, reliable ash-monitoring system suitable for use as a process control sensor. The video-based slurry sensor is mounted in a small sump which is continuously fed with coal tailings. The slurry presentation system uses a pressurized tube to rapidly acquire samples of tailings slurry. The video-based sensor employs a black-and-white television camera to acquire live images of the slurry samples. These images are then processed by the PC-based image analysis system to rapidly determine ash content. An adaptive calibration system is used in conjunction with manual monitoring and sampling to provide a means for continuous improvement of the measurement accuracy. Problems with sample illumination and sample presentation have plagued previous developments of on-line optical sensors. The video-based slurry sensor developed in this work uses a unique sample presentation system to provide high-quality slurry images online. The possibilities of using this technology in other mineral processing applications are abundant. / Master of Science

The advent of multi-photon microscopes and highly sensitive protein sensors enables the recording of astrocyte activities on a large population of cells over a long-time period in vivo. Existing tools cannot fully characterize these activities, both within single cells and at the population-level, because of the insufficiency of current region-of-interest-based approaches to describe the activity that is often spatially unfixed, size-varying, and propagative. Here, we present Astrocyte Quantitative Analysis (AQuA), an analytical framework that releases astrocyte biologists from the ROI-based paradigm. The framework takes an event-based perspective to model and accurately quantify the complex activity in astrocyte imaging datasets, with an event defined jointly by its spatial occupancy and temporal dynamics. To model the signal propagation in astrocyte, we developed graphical time warping (GTW) to align curves with graph-structured constraints and integrated it into AQuA. To make AQuA easy to use, we designed a comprehensive software package. The software implements the detection pipeline in an intuitive step by step GUI with visual feedback. The software also supports proof-reading and the incorporation of morphology information. With synthetic data, we showed AQuA performed much better in accuracy compared with existing methods developed for astrocytic data and neuronal data. We applied AQuA to a range of ex vivo and in vivo imaging datasets. Since AQuA is data-driven and based on machine learning principles, it can be applied across model organisms, fluorescent indicators, experimental modes, and imaging resolutions and speeds, enabling researchers to elucidate fundamental astrocyte physiology. / Doctor of Philosophy / Astrocyte is an important type of glial cell in the brain. Unlike neurons, astrocyte cannot be electrically excited. However, the concentrations of many different molecules inside and near astrocytes change over space and time and show complex patterns. Recording, analyzing, and deciphering these activity patterns enables the understanding of various roles astrocyte may play in the nervous system. Many of these important roles, such as sensory-motor integration and brain state modulation, were traditionally considered the territory of neurons, but recently found to be related to astrocytes. These activities can be monitored in the intracellular and extracellular spaces in either brain slices and living animals, thanks to the advancement of microscopes and genetically encoded fluorescent sensors. However, sophisticated analytical tools lag far behind the impressive capability of generating the data. The major reason is that existing tools are all based on the region-of-interest-based (ROI) approach. This approach assumes the field of view can be segmented to many regions, and all pixels in the region should be active together. In neuronal activity analysis, all pixels in an ROI (region of interest) correspond to a neuron and are assumed to share a common activity pattern (curve). This is not true for astrocyte activity data because astrocyte activities are spatially unfixed, size-varying, and propagative. In this dissertation, we developed a framework called AQuA to detect the activities directly. We designed an accurate and flexible detection pipeline that works with different types of astrocyte activity data sets. We designed a machine learning model to characterize the signal propagation for the pipeline. We also implemented a compressive and user-friendly software package. The advantage of AQuA is confirmed in both simulation studies and three different types of real data sets.

To evaluate repeatability and reproducibility of macular thickness measurements in visually normal eyes using the Topcon 3D OCT-1000. Methods: Phase 1 investigated scan repeatability, the effect of age and pupil dilation. Two groups (6 younger and 6 older participants) had one eye scanned 5 times pre and post- dilation by 1 operator. Phase 2 investigated between-operator, within and between-visit reproducibility. 10 participants had 1 un-dilated eye scanned 3 times on 2 separate visits by 2 operators. Results: Phase 1: No significant difference existed between repeat scans (p=0.75) and no significant difference was found pre- and post-dilation (p=0.54). In the younger group variation was low (95% limits ± 3.62 m) and comparable across all retinal regions. The older group demonstrated greater variation (95% limits ± 7.6 m). Phase 2: For a given retinal location, 95% confidence limits for within-operator, within-visit reproducibility was 5.16 m. This value increased to 5.56 m for the same operator over two visits and to 6.18 m for two operators over two visits. Conclusion: A high level repeatability, close to 6 m, of macular thickness measurement is possible using the 3D OCT- 1000. Measured differences in macular thickness between successive visits that exceed 6 m in pre-presbyopic individuals are therefore likely to reflect actual structural change. OCT measures are more variable in older individuals and it is advisable to take a series of scans so that outliers can be more easily identified.

There is a growing interest in agricultural spray drone (ASD) use for herbicide application in managed turfgrass systems, which historically has precluded aerial application. Considering pesticide deposition accuracy is of utmost importance in managed turfgrass systems, a thorough examination of factors that influence ASD spray deposition patterns is needed. A python-based spray deposition pattern analysis tool, SprayDAT, was developed to estimate spray quality utilizing a cost-effective continuous sampling technique involving digital soand spectrophotometric analysis of blue colorant stains on white Kraft paper. This technique cost 0.2 cents per USD spent on traditional water-sensitive paper (WSP) allowing for continuous sampling necessary for the highly variable deposition patterns delivered by an ASD. SprayDAT conserved droplet densities and more accurately detected stain objects compared to a commonly utilized software, DepositScan, which overestimated stain sizes. However, droplet density exhibited an upper asymptote at 22% stain cover when relating volume median diameter (VMD) due to increasing overlap of stain objects. Spread factor of blue colorant stains was fit to a 2-parameter power equation when compared across six discrete droplet sizes between 112 and 315 µm when droplets were captured in a biphasic solution of polydimethylsiloxane of 100 cSt over 12,500 cSt viscosities. Cumulative digitally assessed stain objects underestimated application volume 270% when compared to the predicted output based on flow rate, coverage, and speed. SprayDAT incorporates a standard curve based on colorant extraction and spectrophotometric analysis to correct this error such that total stain area accurately estimates application volume to within 9%. This relationship between extracted colorant and total stain area, however, is dependent on droplet size spectra. SprayDAT allows users to customize standard curves to address this issue. Using these analysis techniques, continuous sampling of a 29.3-m transect perpendicular to an ASD or ground sprayer spray swath resolved that increasing ASD operational height increases drift and effective swath width while effective application rate, total deposition, and smooth crabgrass control by quinclorac herbicide decreases. Deposition under the ASD was heterogeneous as the coefficient of variation (CV) within the targeted swath exceeded 30% regardless of operational height. At higher operational heights, relative uniformity of spray pattern was improved but droplet density at 11.7 m away from the intended swath edge was up to four times greater and total spray deposited was up to 60% reduced at the highest heights. For each 1-m increase in ASD operational height, 6% of the deposited spray solution, 11% of the effective application rate within the targeted swath, and 7% of smooth crabgrass [Digitaria ischaemum (Schreb.) Schreb. ex Muhl.] population reduction declined. Subsequent studies suggested that total deposition loss with increasing operational height of ASD were likely due to droplet evaporation. Discrete-sized droplets subjected to a 5-m fall in a windless environment exhibited a sigmoidal relationship where 98% volume of 135-µm droplets and approximately 67% volume of 177 – 283 µm diameter droplets evaporated. Addition of drift reduction agents (DRAs) or choosing different nozzle types altered the initial droplet density generated by a flat-fan nozzle. Regardless of DRA additions or nozzle replacement, the distance required to lose 50% of small droplets (< 150 µm diameter) was 6.6 m. Air induction nozzles and DRA admixtures also conserved smooth crabgrass control across 2- and 6-m operational heights, where control was reduced at the 6-m height with a flat fan nozzle without DRA. Spray deposition pattern analysis for multipass ASD and ground applications was conducted by utilizing nighttime UV-fluorescence aerial photography and weed infestation counts in a digitally overlaid grid. Results show that under-application across all devices was consistent and averaged 12%, whereas at least 14% more over-application on the targeted area was observed for ASD, regardless of equipped nozzle types, compared to a ride-on sprayer. Drift also occurred at least 3 times more for ASD application than for a ride-on sprayer and a spray gun sprayer. Using smooth crabgrass infestation annotated from aerial images could not consistently resolve the spatial variability evident in UV-fluorescent imagery presumably due to the innate variability in weed populations. Analysis using SprayDAT revealed insights into factors affecting ASD spray deposition, such as operational height impacting drift, effective swath width, and herbicide efficacy, highlighting the tool's utility in optimizing aerial herbicide applications in turfgrass management. Data suggest that the lowest ASD operational height should be employed to partially mitigate drift and droplet evaporation while improving weed control. Lower operational heights, however, reduce effective swath width and increase heterogeneity of the deposition pattern. Future research should evaluate possible engineering controls for these problems. / Doctor of Philosophy / In recent years, there has been growing interest in using agricultural spray drones (ASD) for applying herbicides in managed turfgrass systems. Traditionally, aerial spraying has not been widely used in these settings, but ASDs are gaining attention. However, there is still a need for a better understanding of how different factors affect spray patterns of ASDs and weed control effectiveness. To address this, novel image analysis software, SprayDAT was developed. It uses white Kraft paper and blue colorant to analyze spray patterns. Compared to traditional methods, SprayDAT provides a cost-effective way to study spray deposition over larger areas, which is important for analyzing the irregular patterns produced by ASDs. The tool showed similar accuracy in detecting spray patterns compared to existing software used with water-sensitive papers, but with some improvements in detecting fine details. SprayDAT was used to analyze spray patterns from ASDs equipped with different nozzles at various heights, as well as ground application methods. It was found that regardless of height, ASDs showed some inconsistency in spray deposition, with about 6% of the spray solution and 11% of the effective application rate being lost for each 1-m increase in ASD height. This loss is likely due to droplet evaporation based on additional laboratory and field studies that directly measured droplet volume loss or stains of small droplets on white paper. In another part of the study, UV-fluorescent nighttime aerial images and weed infestation following herbicide sprays were used to assess spray deposition of multipass ASD applications. It was found that ASDs tended to over-apply in more of the targeted area than ground-based methods and caused more drift of spray to non-target areas. These studies suggest that lower operational heights, such as 2-m above ground, is recommended when controlling weeds with an ASD as effective application rate and weed control will be improved. These low heights, however, increase variability of rate across the intended spray swath and reduce the effective swath width.

<p>Mathematical models of classes of objects can significantly contribute to the analysis of digital images. A major problem in modelling is to establish suitable descriptions that cover not only a single object but also the variation that is usually present within a class of objects.</p><p>The objective of this thesis is to develop more general modelling strategies than commonly used today. In particular, the impact of the human factor in the model creation process should be minimised. It is presumed that the human ability of abstraction imposes undesired constraints on the description. In comparison, common approaches are discussed from the viewpoint of generality.</p><p>The technique considered introduces <i>appearance space </i>as a common framework to represent both shapes and images. In appearance space, an object is represented by a single point in a high-dimensional vector space. Accordingly, objects subject to variation appear as <i>nonlinear manifolds</i> in appearance space. These manifolds are often characterised by only a few intrinsic dimensions. A model of a class of objects is therefore considered equal to the mathematical description of this manifold.</p><p>The presence of nonlinearity motivates the use of artificial auto-associative neural networks in the modelling process. The network extracts nonlinear modes of variation from a number of training examples. The procedure is evaluated on both synthetic and natural data of shapes and images and shows promising results as a general approach to object modelling.</p>

Mathematical models of classes of objects can significantly contribute to the analysis of digital images. A major problem in modelling is to establish suitable descriptions that cover not only a single object but also the variation that is usually present within a class of objects. The objective of this thesis is to develop more general modelling strategies than commonly used today. In particular, the impact of the human factor in the model creation process should be minimised. It is presumed that the human ability of abstraction imposes undesired constraints on the description. In comparison, common approaches are discussed from the viewpoint of generality. The technique considered introduces appearance space as a common framework to represent both shapes and images. In appearance space, an object is represented by a single point in a high-dimensional vector space. Accordingly, objects subject to variation appear as nonlinear manifolds in appearance space. These manifolds are often characterised by only a few intrinsic dimensions. A model of a class of objects is therefore considered equal to the mathematical description of this manifold. The presence of nonlinearity motivates the use of artificial auto-associative neural networks in the modelling process. The network extracts nonlinear modes of variation from a number of training examples. The procedure is evaluated on both synthetic and natural data of shapes and images and shows promising results as a general approach to object modelling.