A Thesis submitted to The University of Arizona College of Medicine - Phoenix in partial fulfillment of the requirements for the Degree of Doctor of Medicine. / Cancer therapies have evolved continuously, with the newest class being immunotherapies targeting the PD‐L1/PD‐1 pathway. This pathway is often overexpressed in malignancies, which allow the aberrant cells to evade the body’s natural immune response that would normally eliminate them. The novel therapies currently being investigated are monoclonal antibodies that target either the PD‐L1 on the tumor cell or the PD‐1 on the lymphocyte. Considering there are significant toxicities with these therapies, namely gastrointestinal and endocrine adverse effects, a predictive tool that could allow physicians which patients are likely to respond to these immunotherapies could spare patients unnecessary therapy and potential economic harm. Since repetitive imaging of patients with cancer is necessary to monitor treatment response, advanced imaging analysis techniques on standard of care images, such as CT scans may provide insights into tumor patterns that could help to predict treatment response. Quantitative texture analysis (QTA) of computed tomography scans has been used in various settings to examine tissue heterogeneity as a predictive biomarker of response; we hypothesized that QTA may have potential value in predicting tumor response to immunotherapy. We performed a QTA on standard of care CT scans from patients to determine if a unique textural imaging signature could be identified that would serve as a predictive biomarker for response to PD‐L1/PD‐1 therapies in subjects with solid tumor malignancies in the lungs, liver, and lymph nodes. This study examined the diagnostic standard of care CT scans of the chest, abdomen, and pelvis (CT CAP) at baseline and follow‐up, which were acquired as part of routine clinical care for tumor staging and treatment response in 20 subjects whose personal health care information was removed prior to analysis. Regions of interest (ROI) were drawn around all identifiable tumor lesions on baseline CT scans provided that tumors were of reasonable size (>10 mm in diameter) and conspicuity. CT texture analysis was performed on these lesions to obtain a histogram readout of tumor texture based upon tissue densities on a per pixel bases. The output values from the QTA platform provided an estimate of tumor signal properties as expressed as the mean pixel density, standard deviation, entropy, kurtosis, skewness, and mean positive pixel values. Each subject was designated as achieving either a RECIST based treatment response or not. Statistical modeling was then conducted using regression techniques. There was no identifiable signature when examining all of the lesions together, but there were statistically significant correlations noted between QTA and RECIST responses for lung‐based lesions. The QTA derived mean pixel density parameter was a major component of separating out responders from non‐response. Of the 14 lung lesions (8 responder vs. 6 nonresponder) there was a significant difference in the mean density with a threshold cutoff of 11.91 (p < 0.0001). A Mann‐Whitney U‐test was performed on the total data set yielding a Z statistic of 2.6 (p=0.0092). Despite the relatively small number of patients in this initial study, there were promising findings regarding the mean density of lesions, suggesting that texture analysis can be used to predict if patients respond to PD‐L1/PD‐1 inhibitors. Further investigation is warranted in a larger population that can be differentiated by tumor type to validate these results.

No description available.

This thesis presents some substantial theoretical analyses and optimal treatments of Kohonen's self-organising map (SOM) algorithm, and explores the practical application potential of the algorithm for vector quantisation, pattern classification, and image processing. It consists of two major parts. In the first part, the SOM algorithm is investigated and analysed from a statistical viewpoint. The proof of its universal convergence for any dimensionality is obtained using a novel and extended form of the Central Limit Theorem. Its feature space is shown to be an approximate multivariate Gaussian process, which will eventually converge and form a mapping, which minimises the mean-square distortion between the feature and input spaces. The diminishing effect of the initial states and implicit effects of the learning rate and neighbourhood function on its convergence and ordering are analysed and discussed. Distinct and meaningful definitions, and associated measures, of its ordering are presented in relation to map's fault-tolerance. The SOM algorithm is further enhanced by incorporating a proposed constraint, or Bayesian modification, in order to achieve optimal vector quantisation or pattern classification. The second part of this thesis addresses the task of unsupervised texture-image segmentation by means of SOM networks and model-based descriptions. A brief review of texture analysis in terms of definitions, perceptions, and approaches is given. Markov random field model-based approaches are discussed in detail. Arising from this a hierarchical self-organised segmentation structure, which consists of a local MRF parameter estimator, a SOM network, and a simple voting layer, is proposed and is shown, by theoretical analysis and practical experiment, to achieve a maximum likelihood or maximum a posteriori segmentation. A fast, simple, but efficient boundary relaxation algorithm is proposed as a post-processor to further refine the resulting segmentation. The class number validation problem in a fully unsupervised segmentation is approached by a classical, simple, and on-line minimum mean-square-error method. Experimental results indicate that this method is very efficient for texture segmentation problems. The thesis concludes with some suggestions for further work on SOM neural networks.

Low back pain is one of the most common and disabling musculoskeletal disorders worldwide and the third most common reason for surgery in the United States. The lower back, or lumbar region, supports most of the body's weight; it controls spinal movement and stability through the interaction between bones, nerves, muscles, ligaments, and fascia within the lumbar region. Any disorder of those tissues could cause low back pain (LBP); emerging evidence indicates that the thoracolumbar fascia (TLF) is the lower back's most pain-sensitive soft tissue structure. TLF consists of dense connective tissue separated by loose connective tissue, allowing TLF layers to pass easily during torso movement. A series of foundational studies found that patients enduring long-term low back pain have different TLF structures than those without LBP. Injuries may result in adhesions and fibrosis, which may cause adjacent dense connective tissue layers to lose independent motion, limiting movement and causing pain. LBP is diagnosed by investigating the patient's medical history to identify symptoms and then examining the patient to determine the cause of the pain. If the pain persists after diagnosis and treatment, further investigation is required; an ultrasound scan is used as the next step. Ultrasound (US) imaging is a non-invasive and instantaneous method to evaluate soft, connective tissue structures such as muscles, tendons, ligaments, and fascia. Even though measuring echo intensity helps evaluate the soft tissues, this method still has limitations in diagnosing LBP; 90 % of all LBP patients are diagnosed with non-specific LBP, referred to as pain with no definitive cause . An in-depth investigation of US images could potentially provide more specificity in identifying sources of LBP. By providing information about soft tissue structure, texture analysis could increase US images' diagnostic power. The texture of an ultrasound image is the variation of pixel intensities throughout the region of interest (ROI) that produces different patterns; texture analysis is an approach that quantifies the characteristic variation of pixel intensities within ROI to describe tissue morphological characteristics. First-order texture analysis, second-order texture analysis, and grey-level run length texture analysis are types of analysis that could be applied to quantify parameters that describe the features of the texture; the grey-level analysis is usually conducted in four directions of the texture. This study has four objectives; the first objective is to use first-order and second-order analysis to determine texture parameters and determine whether those parameters can differentiate between individuals with and without LBP. The second objective is to use grey level run length analysis to quantify texture parameters in four directions (0^°,45^°,90^°,135^°) and examine whether those parameters can differentiate between individuals with and without LBP. The third objective is to determine the correlation between the first, second, and run length parameters. The fourth objective is to explore how first-order, second order and grey level run length parameters are affected by US machine settings. A custom-written MATLAB program was developed to quantify first and second-order texture parameters and grey-level run length parameters. Using JMP software, each parameter was statistically compared between individuals with and without LBP. Among nine first- and second-order texture parameters, four showed statistically significant differences between individuals with and without LBP. Among 44 run-length parameters, 9 showed statistically significant differences between individuals with and without LBP. The current study also revealed some strong correlations between first, second, and run length parameters; it also shows that the US machine setting has minor effects on the three types of parameters. Although the present study was conducted on a relatively small sample size, the results indicate that one direction of grey level run length analysis and first and second-order texture analysis can differentiate between people with and without LBP. / Master of Science / Low back pain (LBP) is one of the most common and disabling musculoskeletal disorders worldwide and the third most common reason for surgery in the United States. Due to LBP's effect on mobility, it is one of the leading causes of absence from work, early retirement, and long-term disability payments. The thoracolumbar fascia (TLF), a connective tissue that stabilizes the trunk, pelvis, and spine, is considered the most sensitive tissue to LBP. LBP diagnosis is based on the patient's medical history to identify symptoms and then on an examination to determine the cause. If the pain persists after diagnosis and treatment, imaging is recommended as the next step. Ultrasound (US) imaging produces a cross-sectional image of the structure and has been used to compare TLF structure in people with and without LBP. Additional analyses must be done to increase US images' ability to diagnose LBP. In the current project, three types of analysis of US images were performed; first-order, second-order, and grey level run length analyses were performed to determine parameters for the images of the two groups of people; selected parameters were noted to distinguish between people with and without LBP.

abstract: Hepatocellular carcinoma (HCC) is a malignant tumor and seventh most common cancer in human. Every year there is a significant rise in the number of patients suffering from HCC. Most clinical research has focused on HCC early detection so that there are high chances of patient's survival. Emerging advancements in functional and structural imaging techniques have provided the ability to detect microscopic changes in tumor micro environment and micro structure. The prime focus of this thesis is to validate the applicability of advanced imaging modality, Magnetic Resonance Elastography (MRE), for HCC diagnosis. The research was carried out on three HCC patient's data and three sets of experiments were conducted. The main focus was on quantitative aspect of MRE in conjunction with Texture Analysis, an advanced imaging processing pipeline and multi-variate analysis machine learning method for accurate HCC diagnosis. We analyzed the techniques to handle unbalanced data and evaluate the efficacy of sampling techniques. Along with this we studied different machine learning algorithms and developed models using them. Performance metrics such as Prediction Accuracy, Sensitivity and Specificity have been used for evaluation for the final developed model. We were able to identify the significant features in the dataset and also the selected classifier was robust in predicting the response class variable with high accuracy. / Dissertation/Thesis / M.S. Industrial Engineering 2013

Patellar Tendinopathy (PT), commonly called "Jumper's Knee", is a condition resulting from repetitive loading of the patellar tendon that presents as anterior knee pain, which is commonly seen in basketball players due to the maneuvers in the sport. Diagnosis of PT often involves a clinical exam followed by ultrasound images for confirmation of the diagnosis to look for key factors of PT. Clinical assessment of ultrasound images of tendons is subjective and requires a high level of experience for reliable interpretation. Thus, there is a need for objective, quantitative methods to assess tendon abnormalities associated with pathology. Ultrasound image texture analysis has emerged as a reliable technique to augment the utility of conventional US imaging, and has recently been shown to distinguish healthy from abnormal tendon and myofascial tissues. The objective of the present study was to conduct image texture analysis to evaluate patellar tendons of collegiate basketball athletes over two seasons. Under an IRB-approved protocol with informed consent, a total of 33 Division 1 collegiate basketball athletes (16 male, 17 female, age 19.9 +/- 1.4 years) underwent clinical evaluation and ultrasound imaging. Four imaging sessions were collected over the course of two years (pre- and post-season). Participants were imaged using a GE LOGIQ S8 (General Electric, USA) ultrasound machine equipped with ML6-15 linear probe. At each imaging session, power Doppler images were collected in the longitudinal and transverse axis, at the proximal, central, and distal regions of the patellar tendon of both knees. Image texture analysis was performed using a custom MATLAB (Mathworks, USA) program to obtain first order (mean, median, variance, skewness, kurtosis, entropy), second order (contrast, energy, and homogeneity), and blob analysis (blob count, BC, and blob area, BA, for 5%, 25%, 50%, 75%, and 95% thresholding values) texture parameters in each image, based upon borders manually drawn by a single researcher. Statistical analysis was conducted to compare imaging sessions (JMP Pro 16, SAS). P-values <0.05 were considered statistically significant. Quantitative texture parameters are able to distinguish characteristics in patellar tendon ultrasound images to distinguish between anatomic region, gender, dominance and pre- to post- season. The 25% and 75% thresholding percentiles effectively showed characteristics of collagen fibers in the patellar tendon. The abnormal diagnosis does not greatly effect texture parameters, which needs to be investigated with more incorporation of grading criteria distinctions and a larger sample size. / Master of Science / Patellar Tendinopathy (PT) is a knee injury that commonly occurs in basketball players. The recovery for PT is often long and the player can still have knee pain when returning to the sport. Diagnosis of PT requires a high level of expertise to consider the patients history, conduct a physical exam and take ultrasound images to look for factors that indicate patellar tendon is damaged. The difficulty of diagnosing PT calls for an objective method to allow for accuracy in assessing patellar tendons. In order to create a more objective measure of ultrasound images, quantitative texture parameters are explored to understand what the brightness values of each pixel and the proximity of pixels together can convey about the image. The objective of this study is to understand what characteristics of the subject (anatomic region, knee dominance, gender, and time point) texture parameters are able to distinguish in patellar tendon ultrasound images.

Texture can be broadly stated as spatial variation of image intensities. Texture analysis and classification is a well researched area for its importance to many computer vision applications. Consequently, much research has focussed on deriving powerful and efficient texture descriptors. Local binary patterns (LBP) and its variants are simple yet powerful texture descriptors. LBP features describe the texture neighbourhood of a pixel using simple comparison operators, and are often calculated based on varying neighbourhood radii to provide multi-resolution texture descriptions. A comprehensive evaluation of different LBP variants on a common benchmark dataset is missing in the literature. This thesis presents the performance for different LBP variants on texture classification and retrieval tasks. The results show that multi-scale local binary pattern variance (LBPV) gives the best performance over eight benchmarked datasets. Furthermore, improvements to the Dominant LBP (D-LBP) by ranking dominant patterns over complete training set and Compound LBP (CM-LBP) by considering 16 bits binary codes are suggested which are shown to outperform their original counterparts. The main contribution of the thesis is the introduction of multi-dimensional LBP features, which preserve the relationships between different scales by building a multi-dimensional histogram. The results on benchmarked classification and retrieval datasets clearly show that the multi-dimensional LBP (MD-LBP) improves the results compared to conventional multi-scale LBP. The same principle is applied to LBPV (MD-LBPV), again leading to improved performance. The proposed variants result in relatively large feature lengths which is addressed using three different feature length reduction techniques. Principle component analysis (PCA) is shown to give the best performance when the feature length is reduced to match that of conventional multi-scale LBP. The proposed multi-dimensional LBP variants are applied for medical image analysis application. The first application is nailfold capillary (NC) image classification. Performance of MD-LBPV on NC images is highest, whereas for second application, HEp-2 cell classification, performance of MD-LBP is highest. It is observed that the proposed texture descriptors gives improved texture classification accuracy.

No description available.

Ovarian follicles in women are fluid-filled structures in the ovary that contain oocytes (eggs). A dominant follicle is physiologically selected and ovulates during the menstrual cycle. We examined the echotexture in ultrasonographic images of the follicle wall of dominant ovulatory follicles in women during natural menstrual cycles and dominant anovulatory follicles which developed in women using oral contraceptives (OC). Texture features of follicle wall regions of both ovulatory and anovulatory dominant follicles were evaluated over a period of seven days before ovulation (natural cycles) or peak estradiol concentrations (OC cycles). Differences in echotexture between the two classes of follicles were found for two co-occurrence matrix derived texture features and two edge-frequency based texture features. Co-occurrence energy and homogeneity were significantly lower for ovulatory follicles while edge density and edge contrast were higher for ovulatory follicles. In the each feature space, the two classes of follicle were adequately separable.</p><p>This thesis employed several statistical approaches to analyses of texture features, such as plotting method and the Mann-Kendall method. A distinct change of feature trend was detected 3 or 4 days before the day of ovulation for ovulatory follicles in the two co-occurrence matrix derived texture features and two edge-frequency-based texture features. Anovulatory follicles, exhibited the biggest variation of the feature value 3 or 4 days before the day on which dominant follicles developed to maximum size. This discovery is believed to correspond to the ovarian follicles responding to system hormonal changes leading to presumptive ovulation.</p>

Ultrasonography is widely used in medical diagnosis with the advantages of being low cost, non-invasive and capable of real time imaging. When interpreting ultrasonographic images of mammalian ovaries, the structures of interest are follicles, corpora lutea (CL) and stroma. This thesis presents an approach to perform CL texture analysis, including detection and segmentation, based on the classiers trained by genetic programming (GP). The objective of CL detection is to determine whether there is a CL in the ovarian images, while the goal of segmentation is to localize the CL within the image. Genetic programming (GP) oers a solution through the evolution of computer programs by methods inspired by the mechanisms of natural selection. Herein, we use rotationally invariant local binary patterns (LBP) to encode the local texture features. These are used by the programs which are manipulated by GP to obtain highly t CL classiers. Grayscale standardization was performed on all images in our data set based on the reference grayscale in each image. CL classication programs were evolved by genetic programming and tested on ultrasonographic images of ovaries. On the bovine dataset, our CL detection algorithm is reliable and robust. The detection algorithm correctly determined the presence or absence of a CL in 93.3% of 60 test images. The segmentation algorithm achieved a mean ( standard deviation) sensitivity and specicity of 0.87 (0.14) and 0.91 (0.05), respectively, over the 30 CL images. Our CL segmentation algorithm is an improvement over the only previously published algorithm, since our method is fully automatic and does not require the placement of an initial contour. The success of these algorithms demonstrates that similar algorithms designed for analysis of in vivo human ovaries are likely viable.