Search

This paper presents an approach for the detection and delineation of the interfaces at the near and far walls of carotid artery using B-mode ultrasound images. After the delineation the system measures automatically the carotid intima-media thickness (IMT) in order to aid the diagnosis of the atherosclerosis. In this method we start by the measurement, at a pixel level, of local features followed by the selection of the most discriminant ones. The next stage is a classification step which assigns a probability to the pixels to belong to an interface, enabling the detection of the carotid artery interfaces. The final artery boundaries are delineated using a dynamic programming approach. The final measurements of IMT produced by the automatic method proposed in this paper were compared with three manual tracings of experts. It was also compared with an automatic method previously developed. The results show that the two automatic detection methods have similar performance, although with slight improvements in the new method, particularly for the the far wall interface.

The Arteriolar-to-Venular Ratio (AVR) is a well known index for the diagnosis of diseases such as diabetes, hypertension or cardiovascular pathologies. This paper presents a fully automatic AVR estimation method which uses a graph-based artery/vein classification approach to classify the retinal vessels by a combination of structural information taken from the vasculature graph with intensity features from the original color image. This method was evaluated on the images of the INSPIRE-AVR dataset. The mean error and the correlation coefficient of obtained results with respect to the reference AVR values were identical to the ones obtained by the second observer using a semi-automated system, which demonstrate the potential of the herein proposed solution for clinical application.

This paper presents a methodology for correcting band distortions in Thin-Layer Chromatography (TLC) images. After the segmentation of image lanes, the intensity profile of each lane column is spatially aligned with a reference profile using a modified version of the Correlation Optimized Warping (COW) algorithm. As the initial partition of the profile into equal length segments proposed by COW can result in the separation of a single band between two segments to be disjointedly aligned, in the proposed method the warping function is only applied to selected profile regions containing groups of adjacent bands. The proposed band correction methodology was assessed using 105 profiles of 105 TLC lanes. A set of features for band characterization was extracted from each lane profile, before and after band distortion correction, and was used as input for three distinct one-class classifiers aiming at band identification. In all cases, the best results of band classification were obtained for the set lanes after band distortion correction.

This paper describes a methodology to remove the background of the lanes in Thin Layer Chromatography (TLC) images, aiming at improving band detection and classification. The storage of the biological samples to be analyzed by TLC is usually done via plastic containers. Filter paper is an alternative that allows reduced costs and higher portability, but it increases the complexity of the image analysis stage due to lane background alteration. In order to overcome this problem, a negative control lane is included in every chromatographic plate. After image preprocessing and lane detection stages, a background profile is generated by processing the negative control lane using the Discrete Wavelet Transform (DWT). This profile is then subtracted to the profiles of all other sample lanes in order to overcome the data degradation introduced by filter paper usage. For assessing the proposed background removal process, 105 TLC lanes, with and without background, were used as input for three one-class classifiers. In all cases, the best results were achieved for the lanes after background removal.

This paper presents an automatic approach for artery/vein (A/V) classification based on the analysis of a graph representing the structure of the retinal vasculature. The entire vascular tree is classified by deciding on the type of each intersection point (graph node) and assigning one of two classes to each vessel segment (graph link). The final label for each vessel segment is accomplished by a combination of structural information taken from the graph (link class) with intensity features measured in the original color image. An accuracy of 88.0% was achieved for the 40 images of the INSPIRE-AVR dataset, thus demonstrating that our method outperforms state-of-the-art approaches for A/V classification.

This paper proposes a method for automating the detection of lanes in Thin-Layer Chromatography images. Our approach includes a preprocessing step to detect the image region of interest, followed by background estimation and removal. This image is then projected onto the horizontal direction to integrate the information into a one-dimensional profile. A smoothing filter is applied to this profile and the outcome is the input of the lane detection process, which is performed in three phases. The first one aims at obtaining an initial set of candidate lanes that are further validated or removed in the second phase. The last phase is a refinement step that allows the inclusion of lanes that are not clearly distinguishable in the profile and that were not included in the initial set. The method was evaluated in 66 chromatography images and achieved values of recall, precision and Fβ-measure of 97.0%, 99.4% and 98.2%, respectively.

Content based image retrieval (CBIR) is employed in medicine to improve radiologists' diagnostic performance. Today effective medical CBIR systems are limited to specific applications, as to reduce the amount of medical knowledge to model. Although supervised approaches could ease the incorporation of medical expertise, its application is not common due to the scarce number of available user annotations. This paper introduces the application of radiology reports to supervise CBIR systems. The concept is to make use of the textual distances between reports to build a transformation in image space through a manifold learning algorithm. A comparison was made between the presented approach and non-supervised CBIR systems, using a Leave-One-Patient-Out evaluation in a database of computer tomography scans of interstitial lung diseases. Supervised CBIR augmented the mean average precision consistently with an increase between 3 to 8 points, which suggests supervision by radiology reports increases CBIR performance.

This paper proposes an automatic method for estimating the location of the optic disc in color images of the retina. The proposed methodology is founded in a new concept, the entropy of vascular directions, which proved to be a reliable measure for assessing the convergence of vessels around an image point. To improve the robustness of the method, the search for the maximum value of entropy is restricted to image areas with high intensity. This new method was evaluated in two publicly available databases, containing both normal and pathological images, and was able to obtain a valid location for the optic disc in 115 out of the 121 images of the two datasets.

In plant development biology, the study of the structure of the plant's root is fundamental for the understanding of the regulation and interrelationships of cell division and cellular differentiation. This is based on the high connection between cell length and progression of cell differentiation and the nuclear state. However, the need to analyse a large amount of images from many replicate roots to obtain reliable measurements motivates the development of automatic tools for root structure analysis. We present a novel automatic approach to detect cell files, the main structure in plant roots, and extract the length of the cells in those files. This approach is based on the detection of local cell file characteristic symmetry using a wavelet based image symmetry measure. The resulting detection enables the automatic extraction of important data on the plant development stage and of characteristics for individual cells. Furthermore, the approach presented reduces in more than 90% the time required for the analysis of each root, improving the work of the biologist and allowing the increase of the amount of data to be analysed for each experimental condition. While our approach is fully automatic a user verification and editing stage is provided so that any existing errors may be corrected. Given five test images it was observed that user did not correct more than 20% of all automatically detected structure, while taking no more than 10% of manual analysis time to do so.

This paper proposes a classification-based method for automating the segmentation of the region of interest (ROI) in digital images of chromatographic plates. Image segmentation is performed in two phases. In the first phase an unsupervised learning method classifies the image pixels into three classes: frame, ROI or unknown. In the second phase, distance features calculated for the members of the three classes are used for deciding on the new label, ROI or frame, for each individual connected segment previously classified as unknown.The segmentation result is post-processed using a sequence of morphological operators beforeobtaining the final ROI rectangular area. The proposed methodology, which is the initial step for the development of a screening tool for Fabry disease, was successfully evaluated in a dataset of 58 chromatographic images.

This paper presents a method for automating the selection of the rejection rate of one-class classifiers aiming at optimizing the classifier performance. These classifiers are used in a new classification approach to deal with class imbalance in Thin-Layer Chromatography (TLC) patterns, which is due to the huge difference between the number of normal and pathological cases, as a consequence of the rarity of Lysosomal Storage Disorders (LSD) diseases. The classification is performed in two decision stages, both implemented using optimized one-class classifiers: the first stage aims at recognizing most of the normal samples; the outliers of this first decision level are presented to the second stage, which is a multiclassifier prepared to deal with both pathological and normal patterns. The results that were obtained proved that the proposed methodology is able to overcome some of the difficulties associated with the choice of the rejection rate of one-class classifiers, and generally contribute to the minimization of the imbalance problem in TLC pattern classification.

The study of cell nuclei is fundamental for plant cell Biology research. To obtain information at cellular level, researchers image cells' nuclei which were modified with fluorescence proteins, through laser scanning confocal microscopy. These images are normally noisy and suffer from high background fluorescence, making grey-scale segmentation approaches inadequate for a usable detection. To obtain a successful detection even at low contrast we investigate the use of a particular convergence filter, the Symmetric Sliding Band filter (SSBF), for cell detection. This filter is based on gradient convergence and not intensity. As such it can detect low contrast cell nuclei which otherwise would be lost in the background noise. Due to the characteristics of cell nuclei morphology, a symmetry constrain is integrated in the filter which corrects some inadequate detections and results in a filter response that is more discriminative. We evaluate the use of this filter for cell nuclei detection on the Arabidopsis thaliana root tip, where the nuclei were stained using yellow fluorescence protein. The resulting cell nuclei detection precision is 89%.

This paper presents a new classification approach to deal with class imbalance in TLC patterns, which is due to the huge difference between the number of normal and pathological cases as a consequence of the rarity of LSD diseases. The proposed architecture is formed by two decision stages: the first is implemented by a one-class classifier aiming at recognizing most of the normal samples; the second stage is a hierarchical classifier which deals with the remaining outliers that are expected to contain the pathological cases and a small percentage of normal samples. We have also evaluated this architecture by a forest of classifiers, using the majority voting as a rule to generate the final classification. The results that were obtained proved that this approach is able to overcome some of the difficulties associated with class imbalance.

In vivoobservation and tracking of the Arabidopsis thalianaroot meristem, by time-lapse confocal microscopy, is important to understand mechanisms like cell division and elongation. The research herein described is based on a large amount of image data, which must be analyzed to determine the location and state of cells. The automation of the process of cell detection/marking is an important step to provide research tools for the biologists in order to ease the search for special events, such as cell division. This paper discusses a hybrid approach for automatic cell segmentation, which selects the best cell candidates from a starting watershed-based image segmentation and improves the result by merging adjacent regions. The selection of individual cells is obtained using a Support Vector Machine (SVM) classifier, based on the shape and edge strength of the cells' contour. The merging criterion is based on edge strength along the line that connects adjacent cells' centroids. The resulting segmentation is largely pruned of badly segmented and over-segmented cells, thus facilitating the study of cell division.

This paper proposes a new approach for the automatic segmentation of the carotid adventitia in longitudinal B-scans, with and without the presence of plaque. The top and bottom adventitia contours are jointly detected with a 3D dynamic programing scheme that searches for the best pair of boundaries according to a specified fuzzy cost function. Some discriminating features of the adventitia in B-mode images are used to reduce the attraction by other edges. The final contours are filtered with a smoothing spline fitting. The proposed approach was quantitatively evaluated in a set of 38 images. In order to avoid high correlation of the results, a maximum of two images was selected from each patient. The carotid boundaries manually traced by a medical expert were used as the ground truth. Several statistics show that the proposed algorithm gives good results in most of the cases, including many poor quality images. Examples of the detected contours are presented and compared with the ground truth.

The aim of this paper is to evaluate and compare the performance of three convergence index (CI) filters when applied to the enhancement of chest radiographs, aiming at the detection of lung nodules. One of these filters, the sliding band filter (SBF), is the proposal of an innovative operator, while the other two CI class members, the iris filter (IF) and the adaptive ring filter (ARF), are already known in this application. To demonstrate the adequacy of the new filter for the enhancement of chest x-ray images, we calculated several figures of merit with the goal of comparing (i) the contrast enhancement capability of the filters, and (ii) the behavior of the filters for the detection of lung nodules. The results obtained for 154 images with nodules of the JSRT database show that the SBF outperforms both the IF and ARF. The proposed filter demonstrated to be a promising enhancement method, thus justifying its use in the first stage of a computer-aided diagnosis system for the detection of lung nodules.

This paper presents an automated method for the selection of a set of lung nodule candidates, which is the first stage of a computer-aided diagnosis system for the detection of pulmonary nodules. An innovative operator, called sliding band filter (SBF), is used for enhancing the lung field areas. In order to reduce the influence of the blood vessels near the mediastinum, this filtered image is multiplied by a mask that assigns to each lung field point an a prioriprobability of belonging to a nodule. The result is further processed with a watershed segmentation method that divides each lung field into a set of non-overlapping areas. Suspicious nodule locations are associated with the regions containing the highest regional maximum values. The proposed method, whose result is an ordered set of lung nodule candidate regions, was evaluated on the 247 images of the JSRT database with very promising results.

In spite of significant advances in image segmentation techniques, evaluation of these methods thus far has been largely subjective. Typically, the effectiveness of a new algorithm is demonstrated only by the presentation of a few segmented images that are evaluated by some method, or it is otherwise left to subjective evaluation by the reader. We propose a new approach for evaluation of segmentation that takes into account not only the accuracy of the boundary localization of the created segments but also the under-segmentation and over-segmentation effects, regardless to the number of regions in each partition. In addition, it takes into account the way humans perceive visual information. This new metric can be applied both to automatically provide a ranking among different segmentation algorithms and to find an optimal set of input parameters of a given algorithm.

The aim of this paper is to examine a multiclassifier approach to the classification of the lung nodules in X-ray chest radiographs. The approach investigated here is based on an image region-based classification whose output is the information of the presence or absence of a nodule in an image region. The classification was made, essentially, in two steps: firstly, a set of rotation invariant features was extracted from the responses of a multi-scale and multi-orientation filter bank; secondly, different classifiers (multi-layer perceptrons) are designed using different features sets and trained in different data. These classifiers are further combined in order to improve the classification performance. The obtained results are promising and can be used for reducing the false-positives nodules detected in a computer-aided diagnosis system.

We present a fully automated method for extracting the lung region from volumetric X-ray CT images based on material decomposition. By modeling the human thorax as a composition of different materials, the proposed method follows a threshold-based, hierarchical voxel classification strategy. The segmentation procedure involves the automatic computation of threshold values and consists on three main steps: patient segmentation and decomposition, large airways extraction and lung parenchyma decomposition, and lung region of interest segmentation. Experimental results were performed on thoracic CT images acquired from 30 patients. The method provides a reproducible set of thresholds for accurate extraction of the lung parenchyma, needed for computer aided diagnosis systems.

The paper presents the methodology developed to solve the class imbalanced problem that occurs in the classification of Thin-Layer Chromatography (TLC) images. The proposed methodology is based on re-sampling, and consists in the undersampling of the majority class (normal class), while the minority classes, which contain Lysosomal Storage Disorders (LSD) samples, are oversampled with the generation of synthetic samples. For image classification two approaches are presented, one based on a hierarchical classifier and another uses a multiclassifier system, where both classifiers are trained and tested using balanced data sets. The results demonstrate a better performance of the multiclassifier system using the balanced sets.

A new algorithm for an effective and automatic segmentation of the carotid wall in ultrasonic images is proposed. It combines the speed of thresholding algorithms with the accuracy, flexibility and robustness of a successful geometric active contour model which incorporates an optimal image segmentation model in a level set framework. Due to the multiphase nature of these images, a sequential minimum cross entropy thresholding is used to get a first approximation of the segments, reducing the problem to a two phase segmentation. This thresholding solution is then used as a starting point for a two phase piecewise constant version of a geometric active contour model to reduce noise, smooth contours, improve their position accuracy and close eventual gaps in the carotid wall.

A method is proposed to segment digital posterior-anterior chest X-ray images. The segmentation is achieved through the registration of a deformable prior model, describing the anatomical structures of interest, to the X-ray image. The deformation of the model is performed using a deformation grid. A coarse matching of the model is done using anatomical landmarks automatically extracted from the image, and maps of oriented edges are used to guide the deformation process, optimized with a probabilistic genetic algorithm. The method is applied to extract the ribcage and delineate the mediastinum and diaphragms. The segmentation is needed for defining the lungs region, used in computer-aided diagnosis systems.

Thin-Layer Chromatography images are used to detect and identify the presence of specific oligosaccharides, expressed by the existence, at different positions, of bands in the gel image. 1D gaussian deconvolution, commonly used for band detection, does not produce good results due to the large curvature observed in the bands. To overcome this uncertainty on the band position, we propose a novel feature extraction methodology that allows an accurate modeling of curved bands. The features are used to classify the data into two different classes, to differentiate normal from pathologic cases. The paper presents the developed methodology together with the analysis and discussion of the results.

Indexado ISI Grouping and segmentation of images remains a challenging problem in computer vision. Recently, a number of authors have demonstrated a good performance on this task using spectral methods that are based on the eigensolution of a similarity matrix. In this paper, we implement a variation of the existing methods that combines aspects from several of the best-known eigenvector segmentation algorithms to produce a discrete optimal solution of the relaxed continuous eigensolution.

We propose a method for segmenting the ribcage boundary of digital postero-anterior chest X-ray images. The segmentation is achieved by first defining image landmarks: the center of the ribcage and, using polar transformation from this point, two initial points belonging to the ribcage. A bank of Gabor filters (in analogy with the simple cells present in the human visual cortex) is used to obtain an orientation edges enhanced image. In this enhanced image, an edge following, starting from the landmarks previously determined, is performed for delineating the left and right sections of the ribcage. The complete segmentation is then accomplished by connecting these sections with the top section of the ribcage, obtained by means of spline interpolation.

Optical flow algorithms generally demand for high computational power and huge storage capacities. This paper is a contribution for real-time implementation of an optical flow algorithm on a pipeline machine. This overall optical flow computation methodology is presented and evaluated on a set of synthetic and real image sequences. Results are compared to other implementations using as measures the average angular error, the optical flow density and the root mean square error. The proposed implementation achieves very low computation delays, allowing operation at standard video frame-rate and resolution. It compares favorably to recent implementations in standard microprocessors and in parallel hardware.

This work aims at developing an automatic method for the analysis of TLC images for measuring a set of features that can be used for the characterization of the distinctive patterns that result from the separation of oligosaccharides contained in human urine. This paper describes the methods developed for the automatic detection of the lanes contained in TLC images, and for the automatic separation of bands for each detected lane. The extraction of quantitative information related with each band was accomplished with two methods: the EM expectation-maximization and nonlinear least squares trust-region algorithms. The results of these methods, as well as additional quantitative information related with each band, are also presented.

The purpose of the research herein presented is the automatic detection of the rib borders in posterior-anterior (PA) digital chest radiographs. In a computer-aided diagnosis system, the precise location of the ribs is important as it allows reducing the false positive in the detection of abnormalities such as nodules, rib lesions and lung lesions. We adopted an edge based approach aiming at detecting the lower border of each rib. For this purpose, the rib geometric model is described as a parabola. For each rib, the upper limit is obtained using the position of the corresponding lower border.

In this paper we derive an upper bound for the average-case generalization error of the mixture of experts modular neural network, based on an average-case generalization error bound for an isolated neural network. By doing this we also generalize a previous bound for this architecture that was restricted to special problems.

The Arabidopsis thaliana is a well defined and a suited system to study plant development at the cellular level. To follow in vivo the root meristem activity under a confocal microscope the image acquisition process was automated through a coherent observation of a fixed point of the root tip. This position information allows the microscope stage control to track the root tip. Root tip estimation is performed following two approaches: computing the root central line intersection with the contour or the maximum filtered contour curvature point. The first method fits the root border with lines, using the Radon transform and a classification procedure. The central line is defined as the line that bisects the angle between these lines. The intersection of the central line with the root contour provides an estimate for the root tip position. The second method is based on contour traversing, followed by convolution of the contour coordinates with a Gaussian kernel. Curvature is computed for this filtered contour. The maximum curvature point provides another root tip estimate. A third method, based on a Kalman estimator is used to select between the previous two outputs. The system allowed the tracking of the root meristem for more than 20 hours in several experiments.

This paper presents a model for the probability of correct classification for the Cooperative Modular Neural Network (CMNN). The model enables the estimation of the performance of the CMNN using parameters obtained from the data set. The performance estimates for the experiments presented are quite accurate (less than 1% relative difference). We compare the CMNN with a multi-layer perceptron with equal number of weights and conclude that the CMNN is preferred for complex problems. We also investigate the error introduced by one of the CMNN voting strategies.

We present a method for detecting the axis of bilateral symmetry in a digital chest X-ray image and subsequently measuring the degree of symmetry of the image. The detection is achieved by analysing rotated-reflected digital chest X-ray images and it is posed as a global optimization problem solved with a probabilistic genetic algorithm (PGA). The global search is initially based on natural peak orientation information related to the orientation of the symmetry axis. Only a few generations of the PGA are needed to achieve convergence to all the images in the database. This method is applied directly on the intensity input image and does not require any prior segmentation.

In this paper we describe the first industrial prototype to characterize automatically the headlamp beam properties using computer vision. The European commission for transportation provides strict regulations that have to be fulfilled as far as headlamp orientations, luminous and geometrical beam properties are concerned. To test the headlamps, the test system has to be properly aligned to the vehicle in order that the measures achieved on brightness and geometrical beam profile can be reliable. The system we present is composed of two integral subsystems. The first consists of a fixed stereo vision system capable of estimating automatically, in real time and with a very high accuracy, the longitudinal axis of the vehicle while it is approaching the stereo rig. The outcome is used to accurately align the second subsystem with respect to the vehicle. This subsystem is composed of a classic optical projection endowed with a CCD camera used to perform automatically radiometric and geometric assessments of the beam projected by the headlamps. Experiments carried out for both the subsystems prove how the high accuracy achieved by our method makes the prototype compliant with current regulations. It is worth remarking that the technology employed is low cost, thus making our approach suitable for commercial headlight tester.