published_or_final_version / Electrical and Electronic Engineering / Doctoral / Doctor of Philosophy

Thermography is a useful tool for analyzing spinal nerve root irritation, but interpretation of digital infrared images is often qualitative and subjective. A new quantitative, computer-aided method for analyzing thermograms, utilizing the human dermatome map, is presented. Image processing and pattern recognition principles needed to accomplish this goal are discussed. Algorithms for segmentation, boundary detection and interpretation of thermograms are presented. An interactive, user-friendly program to perform this analysis has been developed. Due to the relatively large number of images in an exam, speed and simplicity were emphasized in algorithm development. The results obtained correlate well with clinical data and show promise for aiding the diagnosis of spinal nerve root irritation.

Every scheduled treatment at a radiation therapy clinic involves a series of safety protocol to ensure the utmost patient care. Despite safety protocol, on a rare occasion an entirely preventable medical event, an accident, may occur. Delivering a treatment plan to the wrong patient is preventable, yet still is a clinically documented error. This research describes a computational method to identify patients with a novel machine learning technique to combat misadministration.The patient identification program stores face and fingerprint data for each patient. New, unlabeled data from those patients are categorized according to the library. The categorization of data by this face-fingerprint detector is accomplished with new machine learning algorithms based on Sparse Modeling that have already begun transforming the foundation of Computer Vision. Previous patient recognition software required special subroutines for faces and di↵erent tailored subroutines for fingerprints. In this research, the same exact model is used for both fingerprints and faces, without any additional subroutines and even without adjusting the two hyperparameters. Sparse modeling is a powerful tool, already shown utility in the areas of super-resolution, denoising, inpainting, demosaicing, and sub-nyquist sampling, i.e. compressed sensing. Sparse Modeling is possible because natural images are inherrently sparse in some bases, due to their inherrant structure. This research chooses datasets of face and fingerprint images to test the patient identification model. The model stores the images of each dataset as a basis (library). One image at a time is removed from the library, and is classified by a sparse code in terms of the remaining library. The Locally Competetive Algorithm, a truly neural inspired Artificial Neural Network, solves the computationally difficult task of finding the sparse code for the test image. The components of the sparse representation vector are summed by `1 pooling, and correct patient identification is consistently achieved 100% over 1000 trials, when either the face data or fingerprint data are implemented as a classification basis. The algorithm gets 100% classification when faces and fingerprints are concatenated into multimodal datasets. This suggests that 100% patient identification will be achievable in the clinal setting. / Includes bibliography. / Thesis (M.S.)--Florida Atlantic University, 2016. / FAU Electronic Theses and Dissertations Collection

Since the advent of functional magnetic resonance imaging (fMRI) and the rise in popularity of its use for resting state functional connectivity mapping (rs-FCM) to non-invasively detect correlated networks of brain activity in human and animal models, many resting state FCM studies have reported differences in these networks under pathologies such as Alzheimer’s or schizophrenia, highlighting the potential for the method’s diagnostic relevance. A common underlying assumption of this analysis, however, is that the blood oxygen level dependent (BOLD) signal of fMRI is a direct measurement of local neural activity. The BOLD signal is in fact a measurement of the local changes in concentration of deoxy-hemoglobin (HbR). Thus, it is imperative that neurovascular coupling—the relationship between neuronal activity and subsequent hemodynamic activity—be better characterized to enable accurate interpretation of resting state fMRI in the context of clinical usage. This dissertation first describes the development and utility of WFOM paradigm for the robust and easily adaptable imaging of simultaneous neuronal and hemodynamic activity in awake mouse models of health or disease in strains with genetically encoded fluorescent calcium reporters. Subsequent exploration of resting state WFOM data collected in Thy1-GCaMP3 and Thy1-GCaMP6f mouse strains is then presented, namely the characterization of spatiotemporal patterns of neuronal and hemodynamic activity and different modulatory depths of neuronal activity via a toolbox of unsupervised blind source separation (e.g. k-means clustering) and supervised (e.g. non-negative least squares, Pearson correlation) analysis tools. The presence of these different modulatory depths of neuronal activity were then confirmed in another Thy1-jRGECO1a mouse strain using the same imaging scheme. Finally, the dissertation documents the application of the WFOM paradigm and select analysis tools to a novel mouse model of diffusely infiltrating glioma, through which neuronal and hemodynamic activity changes during diffusely infiltrating glioma development which impact temporal coherence of the tumor region activity relative to non-tumor regions activity were recorded and analyzed. The paradigm also allowed for recording of numerous spontaneous occurrences of interictal neuronal activity during which neurovascular coupling is modified in the tumor, as well as occurrences of non-convulsive generalized seizure activity (during which neurovascular is non-linear and cortex eventually suffers hypoxia). The detection of spatiotemporal patterns and different modulatory depths of activity in the awake mouse cortex, as well as observation of changes in functional activity in the context of diffusely infiltrating glioma, provide us with new insights into the possible mechanisms underlying variations in resting state connectivity networks found in resting state fMRI studies comparing health and disease states.

無線膠囊內窺鏡（CE）是一種用於檢查整個胃腸道，尤其是小腸的無創技術。它極大地改善了許多小腸疾病的診斷和管理方式，如不明原因的消化道出血，克羅恩病，小腸腫瘤，息肉綜合征等。儘管膠囊內窺鏡有很好的臨床表現，但它仍然有一定的局限性。主要問題是每次檢查產生約50,000 幅低質量的圖像，對於醫生來說，評估如此大量的圖像是一項非常耗時、耗力的工作。 / 到目前為止，對於膠囊內窺鏡的分析和評估，學者們都把膠囊內窺鏡圖像視為單獨的，獨立的觀測對象。事實並非如此，因為圖像之間往往有顯著的重疊。特別是當膠囊內窺鏡在被小腸蠕動緩緩推動時，它可以捕捉同一病灶的多個視圖。我們的研究目的是使用所有可用的資訊，包括多幅圖像，研究對於膠囊內窺鏡的電腦輔助診斷（CAD）系統。 / 在這篇論文中，我們提出了一個嵌入分類器的多類隱馬爾可夫模型（HMM）的方案，它可以融合多幅相鄰圖像的時間資訊。由於膠囊內窺鏡圖像的品質比較低，我們首先進行預處理，以加強膠囊內窺鏡圖像，增加其對比度，消除噪聲。我們調查研究了多種圖像增強的方法，並調整了它們的參數使其適用於膠囊內窺鏡圖像。 / 對於基於單幅圖像的有監督的分類，AdaBoost 作為一個集成分類器來融合多個分類器，即本論文中的支持向量機（SVM），k-近鄰（k-NN），貝葉斯分類。在分類之前，我們提取和融合了顏色，邊緣和紋理特徵。 / 對於無線膠囊內窺鏡的視頻摘要，我們提出了有監督和無監督的兩類方法。對於有監督方法，我們提出了一個基於隱馬爾可夫模型的，靈活的，可擴展的框架，用於整合膠囊內窺鏡中連續圖像的時間資訊。它可以擴展到多類別，多特徵，多狀態。我們還提出了聯合隱馬爾可夫模型和並行隱馬爾可夫（PHMM）模型對系統進行改進，它們可以被看作是決策級的資訊融合。聯合隱馬爾可夫模型通過多層次的隱馬爾可夫模型，結合不同的資訊來源，對膠囊內窺鏡視頻進行分類和視頻摘要。 並行隱馬爾可夫模型採用貝葉斯推理，在決策時融合多個不同來源的資訊。對於無監督的方法，我們首先提出了一種基於顏色的特徵提取方法。在反色顏色空間中對亮度不變的色度不變矩用來表示膠囊內窺鏡圖像的顏色特徵。接著，我們又提出了一種基於輪廓元（Contourlet）變換的局部二元模式（LBP）作為紋理特徵。在特徵空間中，我們測量了相鄰圖像的距離，並把它視為一個位於二維平面上的開放輪廓上的點。 然後，我們採用一個無參數的關鍵點檢測方法檢測在視頻片段上的突變關鍵點。基於這些突變關鍵點，我們對膠囊內窺鏡視頻進行分割。最後，在每段被分割的視頻片段上，我們通過提取有代表性的關鍵幀來實現膠囊內窺鏡視頻摘要。我們分別用模擬和真實的病人數據進行實驗，對提出的方法進行驗證，結果表明了我們所提出的方案的有效性。它在實現自動評估膠囊內窺鏡圖像上具有很大的潛力。 / Wireless Capsule Endoscopy (CE) is a non-invasive technology to inspect the whole gastrointestinal (GI) tract, especially the small intestine. It has dramatically changed the way of diagnosis and management of many diseases of the small intestine, such as obscure gastrointestinal bleeding, Crohn’s disease, small bowel tumors, polyposis syndromes, etc. Despite its promising clinical findings, it still has some limitations. The main problem is that it requires manual assessment of approximately 50,000 low quality images per examination which is highly time-consuming and labor-intense. / CE analysis and assessment so far treated CE images as individual and independent observations. It is obviously not the case as there is often significant overlap among images. In particular, CE captures multiple views of the same anatomy as the capsule is slowly propelled by peristalsis. Our broader work aims to perform computer aided diagnosis (CAD) in endoscopy using all available information, including multiple images. / In this dissertation, a framework of multi-class Hidden Markov Models (HMM) embedded with statistical classifiers for combining information from multiple CE images is proposed. Due to the low quality of CE image, pre-processing is performed to enhance CE images by increasing the contrast and removing noises. Several image enhancement methods are investigated and customized for CE images. For frame-based supervised classification, AdaBoost is used as the ensemble classifier to combine multiple classifiers, i.e. support vector machine (SVM), k-nearest neighbor (k-NN), and Bayes classifier. Before classification, color, edge and texture features are extracted and fused. Finally, both supervised and unsupervised methods are proposed for CE study synopsis. For supervised method, a flexible and extensible framework based on HMM is developed to integrate temporal information in CE images. It can be extended to multi-class, multi-features, and multi-states. Improvements can be made by combined HMM and Parallel HMM (PHMM) which are introduced as decision-level fusion schemes. Combined HMM considers different sources via a multi-layer HMM model to perform classification and video synopsis. PHMM employs Bayesian inference to combine the recognition results at decision level. For unsupervised method, illumination-independent opponent color moment invariants and local binary pattern (LBP) based on Contourlet transform are explored as color and texture features, respectively. Pair-wise image dissimilarity is measured in the feature space and treated as points on an open contour in a 2-D plane. CE video is segmented based on sudden change points which are detected using a non-parametric key-point detection method. From each segment, representative frames are extracted to summarize the CE video. Validation results on simulated and real patient data show promising performance of the proposed framework. It has great potential to achieve automatic assessment for CE images. / Detailed summary in vernacular field only. / Detailed summary in vernacular field only. / Detailed summary in vernacular field only. / Detailed summary in vernacular field only. / Detailed summary in vernacular field only. / Zhao, Qian. / Thesis (Ph.D.)--Chinese University of Hong Kong, 2012. / Includes bibliographical references (leaves 142-175). / Electronic reproduction. Hong Kong : Chinese University of Hong Kong, [2012] System requirements: Adobe Acrobat Reader. Available via World Wide Web. / Abstract also in Chinese. / Abstract --- p.ii / Acknowledgments --- p.vii / List of Tables --- p.xiii / List of Figures --- p.xv / Chapter 1 --- The Relevance of Synopsis --- p.1 / Chapter 1.1 --- Problem Statement --- p.1 / Chapter 1.2 --- Application - Capsule Endoscopy Assessment --- p.4 / Chapter 1.3 --- Literature Review --- p.9 / Chapter 1.3.1 --- Methods Based on Frame Classification --- p.11 / Chapter 1.3.2 --- Methods Integrating Temporal Information --- p.14 / Chapter 1.4 --- Contributions --- p.19 / Chapter 1.5 --- Organization --- p.23 / Chapter 2 --- Preliminary --- p.25 / Chapter 2.1 --- Hidden Markov Model (HMM) --- p.25 / Chapter 2.2 --- Factorial HMM --- p.35 / Chapter 3 --- Temporal Integration in Capsule Endoscopy Image Analysis --- p.37 / Chapter 3.1 --- Pre-processing --- p.38 / Chapter 3.2 --- Feature Extraction --- p.43 / Chapter 3.3 --- Frame-based Supervised Classification --- p.47 / Chapter 3.3.1 --- Supervised Classification using Individual Frames --- p.47 / Chapter 3.3.2 --- Ensemble Learning Based on AdaBoost --- p.50 / Chapter 3.4 --- Sequence-based Supervised Classification --- p.52 / Chapter 3.5 --- Experiments --- p.58 / Chapter 3.5.1 --- Capsule Endoscopy Image Enhancement --- p.60 / Chapter 3.5.2 --- Frame-based Supervised Classification --- p.67 / Chapter 3.5.3 --- Image Sequence Classification --- p.68 / Chapter 3.6 --- Discussion --- p.80 / Chapter 3.7 --- Summary --- p.82 / Chapter 4 --- Capsule Endoscopy Study Synopsis --- p.98 / Chapter 4.1 --- Supervised Synopsis Using Statistical Models --- p.98 / Chapter 4.2 --- Unsupervised Synopsis via Representative Frame Extraction --- p.100 / Chapter 4.2.1 --- Feature Extraction --- p.100 / Chapter 4.2.2 --- Non-parametric Key-point Detection --- p.111 / Chapter 4.2.3 --- Representative Frame Extraction --- p.112 / Chapter 4.3 --- Experiments --- p.119 / Chapter 4.3.1 --- Supervised Synopsis Based on HMM --- p.119 / Chapter 4.3.2 --- Unsupervised Synopsis --- p.125 / Chapter 4.4 --- Discussion --- p.132 / Chapter 4.5 --- Summary --- p.133 / Chapter 5 --- Conclusions and Future Work --- p.138 / Chapter 5.1 --- Conclusions --- p.138 / Chapter 5.2 --- Future Work --- p.141 / Bibliography --- p.142

Joint detection is vital for characterizing human pose and serves as a foundation for a wide range of computer vision applications such as physical training, health care, entertainment. This dissertation proposed two methods to detect joints in the human body for pose analysis. The first method detects joints by combining body model and automatic feature points detection together. The human body model maps the detected extreme points to the corresponding body parts of the model and detects the position of implicit joints. The dominant joints are detected after implicit joints and extreme points are located by a shortest path based methods. The main contribution of this work is a hybrid framework to detect joints on the human body to achieve robustness to different body shapes or proportions, pose variations and occlusions. Another contribution of this work is the idea of using geodesic features of the human body to build a model for guiding the human pose detection and estimation. The second proposed method detects joints by segmenting human body into parts first and then detect joints by making the detection algorithm focusing on each limb. The advantage of applying body part segmentation first is that the body segmentation method narrows down the searching area for each joint so that the joint detection method can provide more stable and accurate results.

Indiana University-Purdue University Indianapolis (IUPUI) / Mammographic density is an important risk factor for breast cancer, detecting and screening at an early stage could help save lives. To analyze breast density distribution, a good segmentation algorithm is needed. In this thesis, we compared two popularly used segmentation algorithms, EM-MPM and K-means Clustering. We applied them on twenty cases of synthetic phantom ultrasound tomography (UST), and nine cases of clinical mammogram and UST images. From the synthetic phantom segmentation comparison we found that EM-MPM performs better than K-means Clustering on segmentation accuracy, because the segmentation result fits the ground truth data very well (with superior Tanimoto Coefficient and Parenchyma Percentage). The EM-MPM is able to use a Bayesian prior assumption, which takes advantage of the 3D structure and finds a better localized segmentation. EM-MPM performs significantly better for the highly dense tissue scattered within low density tissue and for volumes with low contrast between high and low density tissues. For the clinical mammogram, image segmentation comparison shows again that EM-MPM outperforms K-means Clustering since it identifies the dense tissue more clearly and accurately than K-means. The superior EM-MPM results shown in this study presents a promising future application to the density proportion and potential cancer risk evaluation.

Indiana University-Purdue University Indianapolis (IUPUI) / We present a novel, variational and statistical approach for model-based segmentation. Our model generalizes the Chan-Vese model, proposed for concurrent segmentation of multiple objects embedded in the same image domain. We also propose a novel shape descriptor, namely the Multi-Compartment Distance Functions or mcdf. Our proposed framework for segmentation is two-fold: first, several training samples distributed across various classes are registered onto a common frame of reference; then, we use a variational method similar to Active Shape Models (or ASMs) to generate an average shape model and hence use the latter to partition new images. The key advantages of such a framework is: (i) landmark-free automated shape training; (ii) strict shape constrained model to fit test data. Our model can naturally deal with shapes of arbitrary dimension and topology(closed/open curves). We term our model Active Geometric Model, since it focuses on segmentation of geometric shapes. We demonstrate the power of the proposed framework in two important medical applications: one for morphology estimation of 3D Motor Neuron compartments, another for thickness estimation of Henle's Fiber Layer in the retina. We also compare the qualitative and quantitative performance of our method with that of several other state-of-the-art segmentation methods.