Browsing by Subject "Segmentation"

Now showing 1 - 8 of 8
  • Thumbnail Image
    Item
    Automated segmentation and pathology detection in ophthalmic images
    (2014-07) Roy Chowdhury, Sohini
    Computer-aided medical diagnostic system design is an emerging inter-disciplinary technology that assists medical practitioners for providing quick and accurate diagnosis and prognosis of pathology. Since manual assessments of digital medical images can be both ambiguous and time-consuming, computer-aided image analysis and evaluation systems can be beneficial for baseline diagnosis, screening and disease prioritization tasks. This thesis presents automated algorithms for detecting ophthalmic pathologies pertaining to the human retina that may lead to acquired blindness in the absence of timely diagnosis and treatment. Multi-modal automated segmentation and detection algorithms for diabetic manifestations such as Diabetic Retinopathy and Diabetic Macular Edema are presented. Also, segmentation algorithms are presented that can be useful for automated detection of Glaucoma, Macular Degeneration and Vein Occlusions. These algorithms are robust to normal and pathological images and incur low computationally complexity.First, we present a novel blood vessel segmentation algorithm using fundus images that extracts the major blood vessels by applying high-pass filtering and morphological transforms followed by addition of fine vessel pixels that are classified by a Gaussian Mixture Model (GMM) classifier. The proposed algorithm achieves more than 95% vessel segmentation accuracy on three publicly available data sets. Next, we present an iterative blood vessel segmentation algorithm that initially estimates the major blood vessels, followed by iterative addition of fine blood vessel segments till a novel stopping criterion terminates the iterative vessel addition process. This iterative algorithm is specifically robust to thresholds since it achieves 95.35% vessel segmentation accuracy with 0.9638 area under ROC curve (AUC) on abnormal retinal images from the publicly available STARE data set.We propose a novel rule-based automated optic disc (OD) segmentation algorithm that detects the OD boundary and the location of vessel origin (VO) pixel. This algorithm initially detects OD candidate regions at the intersection of the bright regions and the blood vessels in a fundus image subjected to certain structural constraints, followed by the estimation of a best fit ellipse around the convex hull that combines all the detected OD candidate regions. The centroid of the blood vessels within the segmented OD boundary is detected as the VO pixel location. The proposed algorithm results in an average of 80% overlap score on images from five public data sets.We present a novel computer-aided screening system (DREAM) that analyzes fundus images with varying illumination and fields of view, and generates a severity grade for non-proliferative diabetic retinopathy (NPDR) using machine learning. Initially, the blood vessel regions and the OD region are detected and masked as the fundus image background. Abnormal foreground regions corresponding to bright and red retinopathy lesions are then detected. A novel two-step hierarchical classification approach is proposed where the non-lesions or false positives are rejected in the first step. In the second step, the bright lesions are classified as hard exudates and cotton wool spots, and the red lesions are classified as hemorrhages and micro-aneurysms. Finally, the number of lesions detected per image is combined to generate a severity grade. The DReAM system achieves 100% sensitivity, 53.16% specificity and 0.904 AUC on a publicly available MESSIDOR data set with 1200 images. Additionally, we propose algorithms that detect post-operative laser scars and fibrosed tissues and neovascularization in fundus images. The proposed algorithm achieves 94.74% sensitivity and 92.11% specificity for screening normal images in the STARE data set from the images with proliferative diabetic retinopathy (PDR). Finally, we present a novel automated system that segments six sub-retinal thickness maps from optical coherence tomography (OCT) image stacks of healthy patients and patients with diabetic macular edema (DME). First, each image in the OCT stack is denoised using a Wiener Deconvolution algorithm that estimates the speckle noise variance using a Fourier-domain based structural error. Next, the denoised images are subjected to an iterative multi-resolution high-pass filtering algorithm that detects seven sub-retinal surfaces in six iterative steps. The thicknesses of each sub-retinal layer for all scans from a particular OCT stack are then combined to generate sub-retinal thickness maps. Using the proposed system the average inner sub-retinal layer thickness in abnormal images is estimated as 275 um (r = 0.92) with an average error of 9.3 um, while the average thickness of the outer segments in abnormal images is estimated as 57.4 um (r = 0.74) with an average error of 3.5 um. Further analysis of the thickness maps from abnormal OCT image stacks demonstrates irregular plateau regions in the inner nuclear layer (INL) and outer nuclear layer (ONL), whose area can be estimated with r = 0.99 by the proposed segmentation system.
  • Thumbnail Image
    Item
    Autonomous Navigation On Urban Sidewalks Under Winter Conditions
    (2020-04) Johnson, Reed
    We describe a multi-step approach to facilitate autonomous navigation in snow by small vehicles in urban environments, allowing travel only on sidewalks and paved paths. Our objective is to have a vehicle autonomously navigate from point A on one urban block to point B on another block, crossing from one block to another only at curb-cuts, and stopping when pedestrians get in the way. A small mobile platform is first manually driven along the sidewalks to continuously record LIDAR and Global Navigation Satellite System (GNSS) data when little to no snow is on the ground. Our algorithm automatically post processes the data to generate a labeled traversability map. During this automated process, areas such as grass, sidewalks, stationary obstacles, roads and curb-cuts are identified. By differentiating between these areas using only LIDAR, the vehicle is later able to create a path for travel on only sidewalks or roads and not in other areas. Our localization approach uses an Extended Kalman Filter to fuse the Lightweight and Ground-Optimized LIDAR Odometry and Mapping (LeGO-LOAM) approach with high accuracy GNSS where available, to allow for accurate localization even in areas with poor GNSS, which is often the case in cities and areas covered by tree canopy. This localization approach is used during the data capture stage, prior to the post-processing stage when labeled segmentation is performed, and again during real time autonomous navigation, carried out using the ROS navigation stack. By using LIDAR odometry combined with GNSS, the robot is able to localize under many different weather conditions, including snow and rain, where other algorithms (e.g. AMCL) will likely fail. We were able to successfully have the vehicle autonomously plan and navigate a 1.6km path in an urban snow-covered neighborhood. Our methodology facilitates autonomous navigation functionality under most weather conditions including autonomous wheelchair navigation.
  • Thumbnail Image
    Item
    Characterizing the microvascular branching geometry of the dual blood supply to the liver with micro-CT
    (2013-07) Kline, Timothy Lee
    Microvascular branching geometries determine the efficacy of the transport of nutrients and metabolic products to and from tissues in large-bodied organisms. The general `plan' is that an artery supplies oxygen, nutrients, and hormones to the tissue and a vein removes metabolic products from that tissue. The blood flow to the organ is controlled by the metabolic demand of the organ by a feedback mechanism controlling the arterial lumen diameter. The liver differs from other organs by having two vascular systems delivering its blood - the hepatic artery and the portal vein. The hepatic artery supplies the oxygen needed by liver cells, and the portal vein delivers the molecules absorbed by the gut which need to be processed by the liver tissue for use by other organs in the body. However, how the hepatic artery and portal vein interact is not fully understood in terms of how their relative flows are adjusted, either passively and/or actively, to meet the needs of the liver tissue. This dissertation explores the hypothesis that the hepatic artery's blood mixes with the portal vein's proximal to the hepatic sinusoids (where their mixing is traditionally thought to occur). This is performed utilizing micro-CT to image rat liver lobes injected with a contrast polymer. During the process of exploring this hypothesis, a number of image analysis tools needed to be developed. For one, understanding the level of accuracy by which geometrical measurements can be made by micro-CT is very important because vascular resistance to flow is proportional to the interbranch segment length, as well as inversely proportional to the fourth power of the lumen diameter. Moreover, a single vessel tree contained in a micro-CT image has hundreds, if not thousands of individual interbranch segments and knowledge of the interconnectivity relationship between the segments is important for modeling such properties as pressure distributions and relative blood flow rates. For these reasons, the development of automated measurement methods to measure the length and diameter of interbranch segments and extract the hierarchical structure of vascular trees was performed. These methods were then compared to a gold-standard measurement (obtained by measuring the lengths and diameters of interbranch segments of a microvascular cast by `hand' under a microscope) to understand the level of accuracy obtainable by micro-CT. Having successfully developed accurate automated measurement algorithms (thereby replacing the time-consuming gold standard measurement method), the algorithms were then used to compare and validate other algorithmic approaches, particularly those that quickly extract geometrical information regarding a vascular bed composed of many vessel trees within a micro-CT image. Because the hepatic artery and portal vein are in close proximity to one another as they distribute throughout the liver, the development of a special segmentation method was needed to allow separation of these concomitant vessel systems that may have `false' connections resulting from blurring of the micro-CT image. Finally, an anatomic study of the vasculature of the liver was performed which offered insight into the interaction between the hepatic artery and portal vein. In the case of specimens where only the portal vein was injected with contrast, only the portal vein was opacified, whereas in hepatic artery injections, both the hepatic artery and portal vein were opacified. Also, when different contrast agents were injected into the hepatic artery and the portal vein, the hepatic artery's contrast agent was observed to be mixed in with the different contrast injected into the portal vein. In addition, in high-resolution scans (5$\mu$m cubic voxels) anatomic evidence for hepatic arteriolo-portal venular shunts occurring between the hepatic artery and portal vein branches were found. Simulations were performed in order to rule out the possibility of the observed shunts being artifacts caused by image blurring. Thus, mixing of the hepatic artery and portal vein can occur proximal to the sinusoidal level, and hepatic arteriolo-portal venular shunts may function as a one-way valve-like mechanism, allowing flow only from the hepatic artery to the portal vein (and not the other way around).
  • Thumbnail Image
    Item
    Coupled theoretical and experimental methods to characterize heterogeneous, anisotropic, nonlinear materials: application to cardiovascular tissues
    (2014-10) Witzenburg, Colleen M.
    The Generalized Anisotropic Inverse Mechanics (GAIM) method is able to provide general tissue characteristics in terms of stiffness, anisotropy strength, and preferred orientation. It allows for the computational dissection of samples, capturing regional differences within a single sample nondestructively. However, the linear assumption implicit in GAIM limited its utility, particularly in the case of cardiovascular soft tissues, which exhibit markedly nonlinear behavior when operating at physiologic strain levels. Therefore, GAIM was extended to consider large-deformation kinematics, a nonlinear closed-form structural model of planar fibrous tissue mechanics was utilized to describe the nonlinear behavior of a cardiovascular soft tissue (rat ventricle wall), and the partitioning method utilized by GAIM was replaced with a more robust partitioning scheme. Then, GAIM was applied in a stepwise fashion (NGAIM) in order to capture the full nonlinear kinetics of cardiovascular soft tissues. Finally, experiments characterizing the three-dimensional loading and failure of healthy porcine ascending aorta were discussed. The work presented in this thesis marks the development and use of novel theoretical and experimental approaches for the analysis of complex cardiovascular soft tissues. An analysis method was developed, NGAIM, that can be applied to examine regional mechanical differences in planar, nonlinear, anisotropic, heterogeneous, tissue samples from all over the body which yields full-field stress. Finally, a partnering was proposed which exploits the characterization capacity of NGAIM with the predictive capacity of the multiscale model to create full three-dimensional simulations of cardiovascular soft tissue behavior.
  • Thumbnail Image
    Item
    Quality of Life: Assessment for Transportation Performance Measures
    (Minnesota Department of Transportation, 2013-01) Schneider, Ingrid E.; Guo, Tian; Schroeder, Sierra
    Quality of life (QOL) is a commonly used term. Defining QOL, however, is an ongoing challenge that experts often take on with minimal input from citizens. This groundbreaking research sought citizen input on what comprised QOL and what role transportation played in it. Further, this research explored in detail the important factors across the breadth of transportation and how the Minnesota Department of Transportation (MnDOT) was performing on these important factors. The research encompassed three phases between 2010 and 2011: (1) an extensive literature review on QOL, (2) 24 focus groups that asked Minnesota’s citizens about their QOL, and (3) a mail questionnaire about what matters in quality of life, transportation and their intersection. Eleven related quality of life factors emerged, including transportation: education, employment and finances, environment, housing, family, friends and neighbors, health, local amenities, recreation and entertainment, safety, spirituality/faith/serenity, and transportation. Within transportation, seven important areas were identified that predicted satisfaction with MnDOT services: access, design, environmental issues, maintenance, mobility, safety and transparency. Results reveal that a) QOL is complex and transportation plays an important and consistent role in it across Minnesota; b) transportation is critical to QOL because it connects us to important destinations in aspects that matter most; and c) Minnesotans can readily identify what matters and how the state is performing within the breadth of transportation services.
  • Thumbnail Image
    Item
    Segmentation and Dense Keypoints Estimation of Monkeys
    (2021-12) Yu, Haozheng
    Animal tracking and pose estimation are core topics in neuroscience. However, for monkeys, current deep learning based algorithms often fail to perform well on segmentation and dense keypoints estimation due to the lack of annotated training data. In this thesis, we address this challenge by developing transfer learning based deep learning algorithms without using fully-annotated monkey data. We develop a bootstrapping strategy to refine the pretrained segmentation model on monkey data annotated with 2D sparse landmarks. In addition, we implement a voxel-based visual hull reconstruction approach to recover the 3D monkey pose from the silhouettes. For dense keypoints estimation, we follow a similar bootstrapping strategy to refine a pretrained HRNet, which is then used to learn a dense keypoint detector by leveraging multiview consistency. Our methods outperform the baseline methods on in-the-cage and in-the-wild monkey data.
  • Thumbnail Image
    Item
    Time series segmentation techniques for land cover change detection
    (2013-05) Garg, Ashish
    Ecosystem-related observations from remote sensors on satellites offer a significant possibility for understanding the location and extent of global land cover change. In this study, we focus on time series segmentation techniques in the context of land cover change detection. We propose a model based time series segmentation algorithm inspired by an event detection framework proposed in the field of statistics. We also present a novel model free change detection algorithm for detecting land cover change that is computationally simple, efficient, non-parametric and takes into account the inherent variability present in the remote sensing data. A key advantage of this method is that it can be applied globally for a variety of vegetation without having to identify the right model for specific vegetation types. We evaluate the change detection capacity of the proposed techniques on both synthetic and MODIS EVI data sets. We illustrate the importance and relative ability of different algorithms to account for the natural variation in the EVI data set.
  • Thumbnail Image
    Item
    Towards A Framework For Simultaneous Feature Tracking And Segmentation
    (2016-05) Poling, Bryan
    This is a collection of several works that I have done during my PhD research as a graduate student at the University of Minnesota. There are three parts, each focusing on a different topic in machine learning and computer vision. The common theme underlying these works is the tracking of feature points in motion video and their segmentation by object. Abstracts for each part are included below. Abstract for Part I: We present a novel approach to rigid-body motion segmentation from two views. We use a previously developed nonlinear embedding of two-view point correspondences into a 9-dimensional space and identify the different motions by segmenting lower-dimensional subspaces. In order to overcome mixed and unknown dimensions of subspaces and nonuniform distributions along them we suggest the novel concept of global dimension and its minimization for clustering subspaces with some theoretical motivation. We propose a fast projected gradient algorithm for minimizing global dimension and thus segmenting motions from 2-views. We develop an outlier detection framework around the proposed method, and we present state-of-the-art results on outlier-free and outlier-corrupted two-view data for segmenting motion. Abstract for Part II: In this part, we present a framework for jointly tracking a collection of features in motion video, which enables sharing information between the different features in the scene. Our method exploits the fact that trajectories of features from locally rigid and semi-rigid scenes are approximately confined to low-dimensional subspaces, and it uses this fact to aid the tracking of poor-quality and non-corner-like features by filling in missing information from other, better feature points. Our method significantly improves tracking performance in real-world, poorly lit scenes, does not require explicit modeling of the structure or motion of the scene, and runs in real time on a single CPU core. Abstract for Part III: In this part, we employ low-cost gyroscopes to improve general purpose feature tracking. We use some of the same ideas from Part II, except instead of borrowing information from other features to aid the tracking of poor-quality features, we rely on independent estimates of optical flow from external inertial sensors. Most related previous methods use gyroscopes to initialize and bound the search for features. In contrast, we use them to regularize the tracking energy function so that they can directly assist in the tracking of ambiguous and poor-quality features. We demonstrate that our simple technique offers significant improvements in performance over conventional template-based tracking methods, and is in fact competitive with more complex and computationally expensive state-of-the-art trackers, but at a fraction of the computational cost. Additionally, we show that the practice of initializing template-based feature trackers like KLT (Kanade-Lucas-Tomasi) using gyro-predicted optical flow offers no advantage over using a careful optical-only initialization method, suggesting that some deeper level of integration, like the method we propose, is needed in order to realize a genuine improvement in tracking performance from these inertial sensors.