Search

Pulmonary emphysema is traditionally subcategorized into three subtypes, which have distinct radiological appearances on computed tomography (CT) and can help with the diagnosis of chronic obstructive pulmonary disease (COPD). Automated texture-based quantification of emphysema subtypes has been successfully implemented via supervised learning of these three emphysema subtypes. In this work, we demonstrate that unsupervised learning on a large heterogeneous database of CT scans can generate texture prototypes that are visually homogeneous and distinct, reproducible across subjects, and capable of predicting accurately the three standard radiological subtypes. These texture prototypes enable automated labeling of lung volumes, and open the way to new interpretations of lung CT scans with finer subtyping of emphysema.

Cardiac computed tomography (CT) scans include approximately 2/3 of the lung and can be obtained with low radiation exposure. Large cohorts of population-based research studies reported high correlations of emphysema quantification between full-lung (FL) and cardiac CT scans, using thresholding-based measurements. This work extends a hidden Markov measure field (HMMF) model-based segmentation method for automated emphysema quantification on cardiac CT scans. We show that the HMMF-based method, when compared with several types of thresholding, provides more reproducible emphysema segmentation on repeated cardiac scans, and more consistent measurements between longitudinal cardiac and FL scans from a diverse pool of scanner types and thousands of subjects with ten thousands of scans.

Automated texture analysis of lung computed tomography (CT) images is a critical tool in subtyping pulmonary emphysema and diagnosing chronic obstructive pulmonary disease (COPD). Texton-based methods encode lung textures with nearest-texton frequency histograms, and have achieved high performance for supervised classification of emphysema subtypes from annotated lung CT images. In this work, we first explore characterizing lung textures with sparse decomposition from texton dictionaries, using different regularization strategies, and then extend the sparsity-inducing constraint to the construction of the dictionaries. The methods were evaluated on a publicly available lung CT database of annotated emphysema subtypes. We show that enforcing sparse decompositions from texton dictionaries and unsupervised dictionary learning can achieve high classification accuracy (>90%). The flexibility of sparse-inducing models embedded either in the representation stage or dictionary learning stage has potential in providing consistency in classification performance on heterogeneous lung CT datasets with further parameter tuning.

This paper presents a novel variational segmentation framework combining shape priors and parametric intensity distribution modeling for extracting the fetal envelope on 3D obstetric ultrasound images. To overcome issues related to poor image quality and missing boundaries, we inject three types of information in the segmentation process: tissue-specific parametric modeling of pixel intensities, a shape prior for the fetal envelope and a shape model of the fetus' back. The shape prior is encoded with Legendre moments and used to constraint the evolution of a level-set function. The back model is used to post-process the segmented fetal envelope. Results are presented on 3D ultrasound data and compared to a set of manual segmentations. The robustness of the algorithm is studied, and both visual and quantitative comparisons show satisfactory results obtained by the proposed method on the tested dataset., (Copyright © 2015 Elsevier B.V. All rights reserved.)

Quantitative analysis of positron emission tomography (PET) brain imaging data requires a metabolite-corrected arterial input function (AIF) for estimation of distribution volume and related outcome measures. Collecting arterial blood samples adds risk, cost, measurement error, and patient discomfort to PET studies. Minimally invasive AIF estimation is possible with simultaneous estimation (SIME), but at least one arterial blood sample is necessary. In this study, we describe a noninvasive SIME (nSIME) approach that utilizes a pharmacokinetic input function model and constraints derived from machine learning applied to an electronic health record database consisting of "long tail" data (digital records, paper charts, and handwritten notes) that were collected ancillary to the PET studies. We evaluated the performance of nSIME on 95 [(11)C]DASB PET scans that had measured AIFs. The results indicate that nSIME is a promising alternative to invasive AIF measurement. The general framework presented here may be expanded to other metabolized radioligands, potentially enabling quantitative analysis of PET studies without blood sampling. A glossary of technical abbreviations is provided at the end of this paper.

Quantification of regional cerebral metabolic rate of glucose (rCMRglu) via positron emission tomography (PET) imaging requires measuring the arterial input function (AIF) via invasive arterial blood sampling. In this study we describe a non-invasive approach, the non-invasive simultaneous estimation (nSIME), for the estimation of rCMRglu that considers a pharmacokinetic input function model and constraints derived from machine learning applied to a fusion of individual medical health records and dynamic [(18)F]-FDG-PET brain images data. The results obtained with our data indicate potential for future clinical application of nSIME, with correlation measures of 0.87 for rCMRglu compared to quantification with full arterial blood sampling.

Ultrasonic transient elastography (TE), enables to assess, under active mechanical constraints, the elasticity of the liver, which correlates with hepatic fibrosis stages. This technique is routinely used in clinical practice to assess noninvasively liver stiffness. The Fibroscan system used in this work generates a shear wave via an impulse stress applied on the surface of the skin and records a temporal series of radio-frequency (RF) lines using a single-element ultrasound probe. A shear wave propagation map (SWPM) is generated as a 2-D map of the displacements along depth and time, derived from the correlations of the sequential 1-D RF lines, assuming that the direction of propagation (DOP) of the shear wave coincides with the ultrasound beam axis (UBA). Under the assumption of pure elastic tissue, elasticity is proportional to the shear wave speed. This paper introduces a novel approach to the processing of the SWPM, deriving the maximum likelihood estimate of the shear wave speed when comparing the observed displacements and the estimates provided by the Green's functions. A simple parametric model is used to interface Green's theoretical values of noisy measures provided by the SWPM, taking into account depth-varying attenuation and time-delay. The proposed method was evaluated on numerical simulations using a finite element method simulator and on physical phantoms. Evaluation on this test database reported very high agreements of shear wave speed measures when DOP and UBA coincide.

The behavior and performance of denoising algorithms are governed by one or several parameters, whose optimal settings depend on the content of the processed image and the characteristics of the noise, and are generally designed to minimize the mean squared error (MSE) between the denoised image returned by the algorithm and a virtual ground truth. In this paper, we introduce a new Poisson-Gaussian unbiased risk estimator (PG-URE) of the MSE applicable to a mixed Poisson-Gaussian noise model that unifies the widely used Gaussian and Poisson noise models in fluorescence bioimaging applications. We propose a stochastic methodology to evaluate this estimator in the case when little is known about the internal machinery of the considered denoising algorithm, and we analyze both theoretically and empirically the characteristics of the PG-URE estimator. Finally, we evaluate the PG-URE-driven parametrization for three standard denoising algorithms, with and without variance stabilizing transforms, and different characteristics of the Poisson-Gaussian noise mixture.

Iterative reconstruction with point spread function (PSF) modeling improves contrast recovery in positron emission tomography (PET) images, but also introduces ringing artifacts and over enhancement that is contrast and object size dependent. Mitigation of these artifacts is crucial for clinical and research purposes. In this work we introduce a new iterative regularized reconstruction method that incorporates locally-weighted total variation denoising designed to suppress artifacts induced by PSF modeling. The reconstruction method is evaluated on a simulated cylindrical phantom and preliminary results show that ringing artifacts are suppressed while contrast recovery is maintained.

This paper presents a study of the effects of scanning parameters variability when assessing glioma sizes on MRI. A database of lesions of various shapes and sizes, segmented on 3D-SPGR MRI images, was acquired on 65 patients with low-grade glioma. Simulations of large slice thickness and patient's head rotation were performed, allowing us to study their influence on two size indices: the bi-dimensional diameter product index (computed with the two largest diameters method) and the equivalent diameter index (computed with the volume segmentation method). Results show that thick slices and axial plane rotation can induce average (maximal) uncertainties on the bi-dimensional diameter product index between 32 and 6 % (150 %) for small and large tumors (size range 0.5-286 ml). The uncertainty on the equivalent diameter index, for the same categories of tumors, drops below 8 and 0.1 % (23 %). This study shows that the volume segmentation method is subject to less variability inherent to scanning conditions compared to the two largest diameters method. It also emphasizes the need for strict clinical guidelines on the replication of scanning conditions when performing MRI follow-ups on patients harboring small tumors. These implications await confirmation on a series of real patients being re-scanned with FLAIR MRI.

Over the past two decades, intravascular ultrasound (IVUS) image segmentation has remained a challenge for researchers while the use of this imaging modality is rapidly growing in catheterization procedures and in research studies. IVUS provides cross-sectional grayscale images of the arterial wall and the extent of atherosclerotic plaques with high spatial resolution in real time. In this paper, we review recently developed image processing methods for the detection of media-adventitia and luminal borders in IVUS images acquired with different transducers operating at frequencies ranging from 20 to 45 MHz. We discuss methodological challenges, lack of diversity in reported datasets, and weaknesses of quantification metrics that make IVUS segmentation still an open problem despite all efforts. In conclusion, we call for a common reference database, validation metrics, and ground-truth definition with which new and existing algorithms could be benchmarked.

A differential analysis framework of longitudinal FLAIR MRI volumes is proposed, based on non-linear gray value mapping, to quantify low-grade glioma growth. First, MRI volumes were mapped to a common range of gray levels via a midway-based histogram mapping. This mapping enabled direct comparison of MRI data and computation of difference maps. A statistical analysis framework of intensity distributions in midway-mapped MRI volumes as well as in their difference maps was designed to identify significant difference values, enabling quantification of low-grade glioma growth, around the borders of an initial segmentation. Two sets of parameters, corresponding to optimistic and pessimistic growth estimations, were proposed. The influence and modeling of MRI inhomogeneity field on a novel midway-mapping framework using image models with multiplicative contrast changes was studied. Clinical evaluation was performed on 32 longitudinal clinical cases from 13 patients. Several growth indices were measured and evaluated in terms of accuracy, compared to manual tracing. Results from the clinical evaluation showed that millimetric precision on a specific volumetric radius growth index measurement can be obtained automatically with the proposed differential analysis. The automated optimistic and pessimistic growth estimates behaved as expected, providing upper and lower bounds around the manual growth estimations., (Copyright © 2011. Published by Elsevier B.V.)

Ultrasonic elastography, via vibration-controlled transient elastography (VCTE), enables to assess, under active mechanical constraints, the elasticity of the liver, correlating with fibrosis stages. On the other hand, the same VCTE probe can also be used in passive mode, acquiring RF lines at different locations in the liver. This paper presents a thorough evaluation of passive-mode RF spectral parameters (integrated backscatter coefficient, power spectral index, effective scattering size and spectral variance), for tissue characterization on a large cohort of volunteers with various ranges of elasticity measures. Results showed that capabilities to discriminate between liver and subcutaneous fat tissues were highly variable among spectral parameters. Furthermore, it appears that no in vivo discrimination of liver elasticity/fibrosis stage can be performed with passive RF spectral analysis, at 3.5MHz.

Anatomical models of pregnant women are used in several applications, such as numerical dosimetry, to assess the potential effects of electromagnetic fields on biological tissues, or medical simulation. Recent advances in medical imaging have enabled the generation of realistic and detailed models of human beings. The construction of pregnant woman models remains a complex task, since it is not possible to acquire whole-body images. Only few models have been developed up to now, and they all present some limitations regarding the representation of anatomical variability of the fetus shape and position over the entire gestation. This paper describes a complete methodology that intends to automate each step of the construction of pregnant women models. The proposed approach relies on the segmentation of 3-D ultrasonic and 3-D magnetic resonance imaging (MRI) data, and on dedicated computer graphics tools. The lack of complete anatomical information for the mother in image data is compensated, in an original way, by merging the available information with a synthetic woman model, deformed to match the image-based information. A set of models anatomically validated by clinical experts is presented. They include detailed information on uterofetal units and cover different gestational stages with various fetal positions.

Recent advances in 4D imaging and real-time imaging provide image data with clinically important cardiac dynamic information at high spatial or temporal resolution. However, the enormous amount of information contained in these data has also raised a challenge for traditional image analysis algorithms in terms of efficiency. In this paper, a novel deformable model framework, Active Geometric Functions (AGF), is introduced to tackle the real-time segmentation problem. As an implicit framework paralleling to level-set, AGF has mathematical advantages in efficiency and computational complexity as well as several flexible feature similar to level-set framework. AGF is demonstrated in two cardiac applications: endocardial segmentation in 4D ultrasound and myocardial segmentation in MRI with super high temporal resolution. In both applications, AGF can perform real-time segmentation in several milliseconds per frame, which was less than the acquisition time per frame. Segmentation results are compared to manual tracing with comparable performance with inter-observer variability. The ability of such real-time segmentation will not only facilitate the diagnoses and workflow, but also enables novel applications such as interventional guidance and interactive image acquisition with online segmentation., (Copyright 2009 Elsevier Ireland Ltd. All rights reserved.)

We present a new technique to delineate lumen borders in intravascular ultrasound (IVUS) volumes of images acquired with a high-frequency Volcano (Rancho Cordova, CA) 45MHz transducer. Our technique relies on projection of IVUS sub-volumes onto orthogonal directional brushlet functions. Through selective projection of IVUS sub-volumes images and their Fourier transforms, tissue-specific backscattered magnitudes and phases identified within brushlet coefficients. We take advantage of such characteristics and construct 2.5-dimensional (2.5-D) magnitudes-phase histograms of coefficients in the transformed complex brushlet domain that contain distinct peaks corresponding to blood and non-blood regions. We exploit these peaks to mask out coefficients that represent blood regions and ultimately detect the luminal border after spatial regularization employing a parametric deformable model. We quantify our results by comparing them to manually traced borders by an expert on 2 datasets, containing 108 frames. We show that our approach is well suited for isolating coherent (i.e. plaque) structures from incoherent (i.e. blood) ones in IVUS pullbacks and detecting the lumen border, a challenging problem particularly in images acquired with high frequency transducers.

Vibration-controlled transient elastography (VCTETM) technique is routinely used in clinical practice to assess non-invasively the liver stiffness which is correlated to hepatic fibrosis. Adequate use of the VCTETM probe requires the knowledge of the distance between the skin and the liver parenchyma. This paper compares two methods to estimate this distance using spatial variations of the spectral content of ultrasound radiofrequency (RF) lines, obtained from a probe consisting of a single element ultrasound transducer placed in front of the liver right lobe. Results on a database of 188 patients, including normal-weight and obese persons, show that the spectral variance can accurately discriminate the subcutaneous fat from the liver tissue. The proposed algorithm works in real-time and is suitable for VCTETM scanning protocol setup.

Vascular diseases are among the most important public health problems in developed countries. Given the size and complexity of modern angiographic acquisitions, segmentation is a key step toward the accurate visualization, diagnosis and quantification of vascular pathologies. Despite the tremendous amount of past and on-going dedicated research, vascular segmentation remains a challenging task. In this paper, we review state-of-the-art literature on vascular segmentation, with a particular focus on 3D contrast-enhanced imaging modalities (MRA and CTA). We structure our analysis along three axes: models, features and extraction schemes. We first detail model-based assumptions on the vessel appearance and geometry which can embedded in a segmentation approach. We then review the image features that can be extracted to evaluate these models. Finally, we discuss how existing extraction schemes combine model and feature information to perform the segmentation task. Each component (model, feature and extraction scheme) plays a crucial role toward the efficient, robust and accurate segmentation of vessels of interest. Along each axis of study, we discuss the theoretical and practical properties of recent approaches and highlight the most advanced and promising ones.

Dynamic cardiac metrics, including myocardial strains and displacements, provide a quantitative approach to evaluate cardiac function. However, in current clinical diagnosis, largely 2D strain measures are used despite the fact that cardiac motions are complex 3D volumes over time. Recent advances in 4D ultrasound enable the capability to capture such complex motion in a single image data set. In our previous work, a 4D optical flow based motion tracking algorithm was developed to extract fully 4D dynamic cardiac metrics from such 4D ultrasound data. In order to quantitatively evaluate our tracking method, in-vivo coronary artery occlusion experiments at various locations were performed on three canine hearts. Each dog was screened with 4D ultrasound and sonomicrometry data was acquired during each occlusion study. The 4D ultrasound data from these experiments was then analyzed with our tracking method and estimated principal strain measures were directly compared to those recorded by sonomicrometry, and showed strong agreement. This is the first validation study of optical flow based strain estimation for 4D ultrasound with direct comparison with sonomicrometry using in-vivo data.

Matrix-phased array transducers for real-time 3-D ultrasound enable fast, noninvasive visualization of cardiac ventricles. Typically, 3-D ultrasound images are semiautomatically segmented to extract the left ventricular endocardial surface at end-diastole and end-systole. Automatic segmentation and propagation of this surface throughout the entire cardiac cycle is a challenging and cumbersome task. If the position of the endocardial surface is provided at one or two time frames during the cardiac cycle, automated tracking of the surface over the remaining time frames could reduce the workload of cardiologists and optimize analysis of 3-D ultrasound data. In this paper, we applied a region-based tracking algorithm to track the endocardial surface between two reference frames that were manually segmented. To evaluate the tracking of the endocardium, the method was applied to 40 open-chest dog datasets with 484 frames in total. Ventricular geometry and volumes derived from region-based endocardial surfaces and manual tracing were quantitatively compared, showing strong correlation between the two approaches. Statistical analysis showed that the errors from tracking were within the range of interobserver variability of manual tracing. Moreover, our algorithm performed well on ischemia datasets, suggesting that the method is robust-to-abnormal wall motion. In conclusion, the proposed optical flow-based surface tracking method is very efficient and accurate, providing dynamic "interpolation" of segmented endocardial surfaces.

The presence of non-coherent blood speckle patterns makes the assessment of lumen size in intravascular ultrasound (IVUS) images a challenging problem, especially for images acquired with recent high frequency transducers. In this paper, we present a robust three-dimensional (3D) feature extraction algorithm based on the expansion of IVUS cross-sectional images and pullback directions onto an orthonormal complex brushlet basis. Several features are selected from the projections of low-frequency 3D brushlet coefficients. These representations are used as inputs to a neural network that is trained to classify blood maps on IVUS images. We evaluated the algorithm performance using repeated randomized experiments on sub-samples to validate the quantification of the blood maps when compared to expert manual tracings of 258 frames collected from three patients. Our results demonstrate that the proposed features extracted in the brushlet domain capture well the non-coherent structures of blood speckle, enabling identification of blood pools and enhancement of the lumen area.

We propose a recursive Bayesian model for the delineation of coronary arteries from 3D CT angiograms (cardiac CTA) and discuss the use of discrete minimal path techniques as an efficient optimization scheme for the propagation of model realizations on a discrete graph. Design issues such as the definition of a suitable accumulative metric are analyzed in the context of our probabilistic formulation. Our approach jointly optimizes the vascular centerline and associated radius on a 4D space+scale graph. It employs a simple heuristic scheme to dynamically limit scale-space exploration for increased computational performance. It incorporates prior knowledge on radius variations and derives the local data likelihood from a multiscale, oriented gradient flux-based feature. From minimal cost path techniques, it inherits practical properties such as computational efficiency and workflow versatility. We quantitatively evaluated a two-point interactive implementation on a large and varied cardiac CTA database. Additionally, results from the Rotterdam Coronary Artery Algorithm Evaluation Framework are provided for comparison with existing techniques. The scores obtained are excellent (97.5% average overlap with ground truth delineated by experts) and demonstrate the high potential of the method in terms of robustness to anomalies and poor image quality.

This paper presents the implementation and quantitative evaluation of a multiphase three-dimensional deformable model in a level set framework for automated segmentation of brain MRIs. The segmentation algorithm performs an optimal partitioning of three-dimensional data based on homogeneity measures that naturally evolves to the extraction of different tissue types in the brain. Random seed initialization was used to minimize the sensitivity of the method to initial conditions while avoiding the need for a priori information. This random initialization ensures robustness of the method with respect to the initialization and the minimization set up. Postprocessing corrections with morphological operators were applied to refine the details of the global segmentation method. A clinical study was performed on a database of 10 adult brain MRI volumes to compare the level set segmentation to three other methods: "idealized" intensity thresholding, fuzzy connectedness, and an expectation maximization classification using hidden Markov random fields. Quantitative evaluation of segmentation accuracy was performed with comparison to manual segmentation computing true positive and false positive volume fractions. A statistical comparison of the segmentation methods was performed through a Wilcoxon analysis of these error rates and results showed very high quality and stability of the multiphase three-dimensional level set method.

Protein production has experienced great advances in recent years. In particular, high throughput protein production, coupled with the use of robotics, outputs thousands of mixtures in micro-array wells. To detect the presence of protein crystal formation, images of these wells are acquired regularly using robotic cameras. Traditionally, a crystallographer would manually process each image--identifying the wells that resulted in protein crystal formation. This manual inspection process is slow and given the high rate of mixture output, it has become near impossible for crystallographers keep up. Our aim is to create an automated method of detecting which wells have crystals and which ones do not. We make use of a neural network that is trained based on manually classified ground truth data. After it is trained, the automatic classifier would give a binary output --a value of one for the detection of crystals and precipitates in images and a value of zero otherwise. In our previous papers, the core methods of using multi-scale Laplacian image representation to extract image features and the implementation of the neural network classifier were discussed. Here we present a new, optimized approach to training the neural network and results from a large-scale test. We claim that the neural network can be better trained if the training image dataset is optimized in the sense that ambiguous images are removed during the initial training processes. Incremental training is implemented so that the network can be improved as more data becomes available. From initial results with training based on a 6,000 optimized image dataset, the accuracy of the improved classifier approaches 95% in identifying a wide array of images.

Quantitative analysis of cardiac motion is of great clinical interest in assessment of ventricular function. Ultrasound imaging, especially matrix transducers acquiring real-time three dimensional data, provide valuable information, from which quantitative measures of cardiac function can be extracted via optical flow computation. Such analysis requires tracking of the image brightness patterns with underlying assumptions of visual persistency. We present a review of myocardial motion analysis on cardiac ultrasound, based on optical-flow computation, with phantom and clinical evaluations for segmental wall assessment and motion features quantification.

Among screening modalities, echocardiography is the fastest, least expensive and least invasive method for imaging the heart. A new generation of three-dimensional (3-D) ultrasound (US) technology has been developed with real-time 3-D (RT3-D) matrix phased-array transducers. These transducers allow interactive 3-D visualization of cardiac anatomy and fast ventricular volume estimation without tomographic interpolation as required with earlier 3-D US acquisition systems. However, real-time acquisition speed is performed at the cost of decreasing spatial resolution, leading to echocardiographic data with poor definition of anatomical structures and high levels of speckle noise. The poor quality of the US signal has limited the acceptance of RT3-D US technology in clinical practice, despite the wealth of information acquired by this system, far greater than with any other existing echocardiography screening modality. We present, in this work, a clinical study for segmentation of right and left ventricular volumes using RT3-D US. A preprocessing of the volumetric data sets was performed using spatiotemporal brushlet denoising, as presented in previous articles Two deformable-model segmentation methods were implemented in 2-D using a parametric formulation and in 3-D using an implicit formulation with a level set implementation for extraction of endocardial surfaces on denoised RT3-D US data. A complete and rigorous validation of the segmentation methods was carried out for quantification of left and right ventricular volumes and ejection fraction, including comparison of measurements with cardiac magnetic resonance imaging as the reference. Results for volume and ejection fraction measurements report good performance of quantification of cardiac function on RT3-D data compared with magnetic resonance imaging with better performance of semiautomatic segmentation methods than with manual tracing on the US data.

Expert hand-drawing of magnetic resonance image (MRI) features can be tedious and time consuming. MRI of the knee were acquired from eight subjects to develop an automated segmentation approach. The regions of interest (ROI) were femur, tibia, and patella cartilage. The Karhunen-Loeve transformation was used to construct prototypical ROI with accentuated features and reduced noise level. Adaptive template matching was then used to translate the prototypical ROI locations for detection and optimal overlap of ROI in test images. Cartilage boundaries at the optimal overlap area were computed based on standard gradient methods.