This PDF file contains the front matter associated with SPIE Proceedings Volume 8000, including the Title Page, Copyright information, Table of Contents, Introduction, and Conference Committee listing.

The optical ow enables the accurate estimation of cardiac motion. In this research, a sparse based algorithm is used to estimate the optical ow in cardiac magnetic resonance images. The dense optical ow eld is represented using a discrete cosine basis dictionary aiming at a sparse representation. The optical ow is estimated in this transformed space by solving a L1 problem inspired on compressive sensing techniques. The algorithm is validated using four synthetic image sequences whose velocity eld is known. A comparison is performed with respect to the Horn and Schunck and the Lucas and Kanade algorithm. Then, the technique is applied to a magnetic resonance image sequence. The results show average magnitude errors as low as 0.35 % for the synthetic sequences, however results on real data are not consistent with respect to the obtained by other algorithms suggesting the need for additional constrains for coping with the dense noise.

In this work is presented a novel strategy that tracks the right ventricle (RV) shape during a whole cardiac cycle in magnetic resonance sequences (MRC). The proposed approach obtains a set of spatio-temporal observations from a bidirectional per pixel motion descriptor which are each time fused with prior learned edges. A main advantage of the proposed approach is a robust MRI heart characterization that is regularized by a prior information, obtaining in each cardiac state coherent results. The proposed approach achieves a Dice Score of 0.64 evaluated over 16 patients.

Assessment of the cardiac Left Ventricle (LV) wall motion is generally based on visual inspection or quantitative analysis of 2D+t sequences acquired in short-axis cardiac cine-Magnetic Resonance Imaging (MRI). Most often, cardiac dynamic is globally analized from two particular phases of the cardiac cycle. In this paper, we propose an automated method to classify regional wall motion in LV function based on spatio-temporal pro les and Support Vector Machines (SVM). This approach allows to obtain a binary classi cation between normal and abnormal motion, without the need of pre-processing and by exploiting all the images of the cardiac cycle. In each short- axis MRI slice level (basal, median, and apical), the spatio-temporal pro les are extracted from the selection of a subset of diametrical lines crossing opposites LV segments. Initialized at end-diastole phase, the pro les are concatenated with their corresponding projections into the succesive temporal phases of the cardiac cycle. These pro les are associated to di erent types of information that derive from the image (gray levels), Fourier, Wavelet or Curvelet domains. The approach has been tested on a set of 14 abnormal and 6 healthy patients by using a leave-one-out cross validation and two kernel functions for SVM classi er. The best classi cation performance is yielded by using four-level db4 wavelet transform and SVM with a linear kernel. At each slice level the results provided a classi cation rate of 87.14% in apical level, 95.48% in median level and 93.65% in basal level.

The acquisition of ECG-gated cine magnetic resonance images of the heart is routinely performed in apnea in order to suppress the motion artifacts caused by breathing. However, many factors including the 2D nature of the acquisition and the use of di erent beats to acquire the multiple-view cine images, cause this kind of artifacts to appear. This paper presents the qualitative evaluation of a method aiming to remove motion artifacts in multipleview cine images acquired on patients with hypertrophic cardiomyopathy diagnosis. The approach uses iconic registration to reduce for in-plane artifacts in long-axis-view image stacks and in-plane and out-of-plane motion artifacts in sort-axis-view image stack. Four similarity measures were evaluated: the normalized correlation, the normalized mutual information, the sum of absolute voxel di erences and the Slomka metric proposed by Slomka et al. The qualitative evaluation assessed the misalignment of di erent anatomical structures of the left ventricle as follows: the misalignment of the interventricular septum and the lateral wall for short-axis-view acquisitions and the misalignment between the short-axis-view image and long-axis-view images. Results showed the correction using the normalized correlation as the most appropriated with an 80% of success.

This article describes a perceptual approach to calculate the optical flow estimation of the left ventricle in a short axis view of the heart in computer tomography images. The method is based on the the Hermite transform which is an image representation model that incorporates some of the more important properties of the first stages of the human visual system. Our optical flow estimation approach incorporates a differential approach that uses the steered Hermite coefficients as local constraints and uses the implicit multiresolution scheme of the Hermite transform to compute large displacements. It also involves several of the constraints seen in the current differential methods which allows obtaining an accurate optical flow. We use the anatomic short axis view of the heart to calculate the optical flow estimation instead of the original CT images of the axial plane. This view allows visualizing the left ventricle like a circular structure, which is more suitable for visualization of the left ventricle motion.

In this paper, lumped-parameter models of the cardiovascular system, the cardiac electrical conduction system and a pacemaker are coupled to generate mitral ow pro les for di erent atrio-ventricular delay (AVD) con gurations, in the context of cardiac resynchronization therapy (CRT). First, we perform a local sensitivity analysis of left ventricular and left atrial parameters on mitral ow characteristics, namely E and A wave amplitude, mitral ow duration, and mitral ow time integral. Additionally, a global sensitivity analysis over all model parameters is presented to screen for the most relevant parameters that a ect the same mitral ow characteristics. Results provide insight on the in uence of left ventricle and atrium in uence on mitral ow pro les. This information will be useful for future parameter estimation of the model that could reproduce the mitral ow pro les and cardiovascular hemodynamics of patients undergoing AVD optimization during CRT.

In structural Magnetic Resonance Imaging (MRI), neurodegenerative diseases generally present complex brain patterns that can be correlated with di erent clinical onsets of this pathologies. An objective method that aims to determine both global and local changes is not usually available in clinical practice, thus the interpretation of these images is strongly dependent on the radiologist's skills. In this paper, we propose a strategy which interprets the brain structure using a framework that highlights discriminant brain patterns for neurodegenerative diseases. This is accomplished by combining a probabilistic learning technique, which identi es and groups regions with similar visual features, with a visual saliency method that exposes relevant information within each region. The association of such patterns with a speci c disease is herein evaluated in a classi cation task, using a dataset including 80 Alzheimer's disease (AD) patients and 76 healthy subjects (NC). Preliminary results show that the proposed method reaches a maximum classi cation accuracy of 81.39%.

The Default Mode Network (DMN) is a resting state network widely used for the analysis and diagnosis of mental disorders. It is normally detected in fMRI data, but for its detection in data corrupted by motion artefacts or low neuronal activity, the use of a robust analysis method is mandatory. In fMRI it has been shown that the signal-to-noise ratio (SNR) and the detection sensitivity of neuronal regions is increased with di erent smoothing kernels sizes. Here we propose to use a multiscale decomposition based of a linear scale-space representation for the detection of the DMN. Three main points are proposed in this methodology: rst, the use of fMRI data at di erent smoothing scale-spaces, second, detection of independent neuronal components of the DMN at each scale by using standard preprocessing methods and ICA decomposition at scale-level, and nally, a weighted contribution of each scale by the Goodness of Fit measurement. This method was applied to a group of control subjects and was compared with a standard preprocesing baseline. The detection of the DMN was improved at single subject level and at group level. Based on these results, we suggest to use this methodology to enhance the detection of the DMN in data perturbed with artefacts or applied to subjects with low neuronal activity. Furthermore, the multiscale method could be extended for the detection of other resting state neuronal networks.

The detection, segmentation and quantification of multiple sclerosis (MS) lesions on magnetic resonance images (MRI) has been a very active field for the last two decades because of the urge to correlate these measures with the effectiveness of pharmacological treatment. A myriad of methods has been developed and most of these are non specific for the type of lesions and segment the lesions in their acute and chronic phases together. On the other hand, radiologists are able to distinguish between several stages of the disease on different types of MRI images. The main motivation of the work presented here is to computationally emulate the visual perception of the radiologist by using modeling principles of the neuronal centers along the visual system. By using this approach we are able to detect the lesions in the majority of the images in our population sample. This type of approach also allows us to study and improve the analysis of brain networks by introducing a priori information.

This paper presents a fully automatic method that condenses relevant morphometric information from a database of magnetic resonance images (MR) labeled as either normal (NC) or Alzheimer's disease (AD). The proposed method generates class templates using Nonnegative Matrix Factorization (NMF) which will be used to develop an NC/AD classi cator. It then nds regions of interest (ROI) with discerning inter-class properties. by inspecting the di erence volume of the two class templates. From these templates local probability distribution functions associated to low level features such as intensities, orientation and edges within the found ROI are calculated. A sample brain volume can then be characterized by a similarity measure in the ROI to both the normal and the pathological templates. These characteristics feed a simple binary SVM classi er which, when tested with an experimental group extracted from a public brain MR dataset (OASIS), reveals an equal error rate measure which is better than the state-of-the-art tested on the same dataset (0:9 in the former and 0:8 in the latter).

Surface electromyographic (SEMG) signals are commonly used as control signals in prosthetic and orthotic devices. Super cial electrodes are placed on the skin of the subject to acquire its muscular activity through this signal. The muscle contraction episode is then in charge of activating and deactivating these devices. Nevertheless, there is no gold standard" to detect muscle contraction, leading to delayed responses and false and missed detections. This fact motivated us to propose a new approach that compares a smoothed version of the SEMG signal with a xed threshold, in order to detect muscle contraction episodes. After preprocessing the SEMG signal, the smoothed version is obtained using a moving average lter, where three di erent window lengths has been evaluated. The detector was tuned by maximizing sensitivity and speci city and evaluated using SEMG signals obtained from the anterior tibial and gastrocnemius muscles, taken during the walking of ve subjects. Compared with traditional detection methods, we obtain a reduction of 3 ms in the detection delay, an increase of 8% in sensitivity but a decrease of 15% in speci city. Future work is directed to the inclusion of a temporal threshold (a double-threshold approach) to minimize false detections and reduce detection delays.

Today, it is generally accepted that current methods for biophysical antenatal surveillance do not facilitate a comprehensive and reliable assessment of foetal well-being and thus, that continuing research into alternative methods is necessary to improve antenatal monitoring procedures. Here, attention has been paid to the abdominal phonogram, a signal that is recorded by positioning an acoustic sensor on the maternal womb and contains valuable information about foetal status, but which is hidden by maternal and environmental sources. To recover such information, this work describes single-channel independent component analysis (SCICA) as an alternative signal processing approach for analyzing the abdominal phonogram. The approach, based on the method of delays, the Temporal Decorrelation Source SEParation implementation (TDSEP) of Independent Components Analysis (ICA), and an automatic grouping algorithm, has managed to successfully retrieve estimates of: (1) the foetal cardiac activity (in the form of the foetal phonocardiogram, FPCG), (2) the maternal cardiovascular activity (in the form of the maternal phonocardiogram, MPCG, and/or pulse wave), (3) the maternal respiratory activity (in the form of the maternal respirograma, MResp), and (4) noise (N). These results have been obtained from a dataset of 25 single-channel phonograms and point at the possibilities of using SCICA to address a fundamental problem faced in antenatal surveillance, i.e. the extraction of information from a non-invasive signal like the abdominal phonogram. Future work will test the possibility of using SCICA to recover information regarding the foetal breathing movements (FBM), another physiological parameter of interest in foetal surveillance.

In the ongoing investigation of integrating Knowledge Discovery in Databases (KDD) into neuroscience, we present a paper that facilitates overcoming the two challenges preventing this integration. Pathological oscillations found in the human brain are difficult to evaluate because 1) there is often no time to learn and train off of the same distribution in the fatally sick, and 2) sinusoidal signals found in the human brain are complex and transient in nature requiring large data sets to work with which are costly and often very expensive or impossible to acquire. Overcoming these challenges in today's neuro-intensive-care unit (ICU) requires insurmountable resources. For these reasons, optimizing KDD for pathological oscillations so machine learning systems can predict neuropathological states would be of immense value. Domain adaptation, which allows a way of predicting on a separate set of data than the training data, can theoretically overcome the first challenge. However, the challenge of acquiring large data sets that show whether domain adaptation is a good candidate to test in a live neuro ICU remains a challenge. To solve this conundrum, we present a methodology for generating synthesized neuropathological oscillations for domain adaptation.

Computer Assisted Orthopedic Surgery (CAOS) requires a correct registration between the patient in the operating room and the virtual models representing the patient in the computer. In order to increase the precision and accuracy of the registration a set of new techniques that eliminated the need to use fiducial markers have been developed. The majority of these newly developed registration systems are based on costly intraoperative imaging systems like Computed Tomography (CT scan) or Magnetic resonance imaging (MRI). An alternative to these methods is the use of an Ultrasound (US) imaging system for the implementation of a more cost efficient intraoperative registration solution. In order to develop the registration solution with the US imaging system, the bone surface is segmented in both preoperative and intraoperative images, and the registration is done using the acquire surface. In this paper, we present the a preliminary results of a new approach to segment bone surface from ultrasound volumes acquired by means 3D freehand ultrasound. The method is based on the enhancement of the voxels that belongs to surface and its posterior segmentation. The enhancement process is based on the information provided by eigenanalisis of the multiscale 3D Hessian matrix. The preliminary results shows that from the enhance volume the final bone surfaces can be extracted using a singular value thresholding.

In this work, we present a novel fully automated multimodal elastic registration method for medical images. The new methodology combines a novel mapping based on the quantification of the intensity uncertainty of the neighborhood pixels, with a monomodal fluid like registration technique; thus the methodology can be summarized as a two-step technique. First, a mapping over both multimodal images is applied. This mapping provides information about the intensity uncertainty of the neighborhood pixels in both images, and it is based on the entropy computed over a local region. Second, a monomodal non-rigid registration is achieved between the transformed images. For this step, it is proposed to use a registration based on fluid-models: demons, diffeomorphic-demons, and a variation of the classical optical-flow. To evaluate the algorithm, a set composed by 12 magnetic resonance images of different modalities (T1, T2 and proton density) were taken from a brain model, and these images were modified by a set of controlled elastic deformations (using splines), in order to generate ground-truths to be registered with the proposed technique. The obtained results in this work showed an average error of less than 1.3 mm by combining the local uncertainty mapping with the diffeomorphic-demons technique, suggesting that the proposed methodology could be considered as a new alternative for fully automated multimodal non-rigid registrations on medical applications, which also ensures to obtain only possible physically deformations.

A new wavelet based algorithm for performing denoising of multiple frame Optical Coherence Tomography (OCT) data is proposed. The method is based on the assumption that noise between frames is uncorrelated. By comparing multiple frames using an appropriate similarity measure we can distinguish between unwanted noise and image features. Two similarity measures are used for weighting wavelet coefficients of each frame. Final denoised image is constructed from weighted and averaged frames. Quantitative and qualitative analysis reveal the superiority of proposed algorithm over existing denoising approaches.

Architectural distortion is a breast cancer sign, characterized by spiculated patterns that define the disease malignancy level. In this paper, the radial spiculae of a typical architectural distortion were characterized by a new strategy. Firstly, previously selected Regions of Interest are divided into a set of parallel and disjoint bands (4 pixels the ROI length), from which intensity integrals (coefficients) are calculated. This partition is rotated every two degrees, searching in the phase plane the characteristic radial spiculation. Then, these coefficients are used to construct a fully connected graph whose edges correspond to the integral values or coefficients and the nodes to x and y image positions. A centrality measure like the first eigenvector is used to extract a set of discriminant coefficients that represent the locations with higher linear energy. Finally, the approach is trained using a set of 24 Regions of Interest obtained from the MIAS database, namely, 12 Architectural Distortions and 12 controls. The first eigenvector is then used as input to a conventional Support Vector Machine classifier whose optimal parameters were obtained by a leave-one-out cross validation. The whole method was assessed in a set of 12 RoIs with different distribution of breast tissues (normal and abnormal), and the classification results were compared against a ground truth, already provided by the data base, showing a precision rate of 0.583%, a sensitivity rate of 0.833% and a specificity rate of 0.333%.

Image-guided interventions allow the physician to have a better planning and visualization of a procedure. 3D freehand ultrasound is a non-invasive and low-cost imaging tool that can be used to assist medical procedures. This tool can be used in the diagnosis and treatment of breast cancer. There are common medical practices that involve large needles to obtain an accurate diagnosis and treatment of breast cancer. In this study we propose the use of 3D freehand ultrasound for planning and guiding such procedures as core needle biopsy and radiofrequency ablation. The proposed system will help the physician to identify the lesion area, using image-processing techniques in the 3D freehand ultrasound images, and guide the needle to this area using the information of position and orientation of the surgical tools. We think that this system can upgrade the accuracy and efficiency of these procedures.

Nowadays, the de nition of patient-speci c constraints in prostate cancer radiotherapy planning are solely based on dose-volume histogram (DVH) parameters. Nevertheless those DVH models lack of spatial accuracy since they do not use the complete 3D information of the dose distribution. The goal of the study was to propose an automatic work ow to de ne patient-speci c rectal sub-regions (RSR) involved in rectal bleeding (RB) in case of prostate cancer radiotherapy. A multi-atlas database spanning the large rectal shape variability was built from a population of 116 individuals. Non-rigid registration followed by voxel-wise statistical analysis on those templates allowed nding RSR likely correlated with RB (from a learning cohort of 63 patients). To de ne patient-speci c RSR, weighted atlas-based segmentation with a vote was then applied to 30 test patients. Results show the potentiality of the method to be used for patient-speci c planning of intensity modulated radiotherapy (IMRT).

This paper presents a novel method for basal-cell carcinoma detection, which combines state-of-the-art methods for unsupervised feature learning (UFL) and bag of features (BOF) representation. BOF, which is a form of representation learning, has shown a good performance in automatic histopathology image classi cation. In BOF, patches are usually represented using descriptors such as SIFT and DCT. We propose to use UFL to learn the patch representation itself. This is accomplished by applying a topographic UFL method (T-RICA), which automatically learns visual invariance properties of color, scale and rotation from an image collection. These learned features also reveals these visual properties associated to cancerous and healthy tissues and improves carcinoma detection results by 7% with respect to traditional autoencoders, and 6% with respect to standard DCT representations obtaining in average 92% in terms of F-score and 93% of balanced accuracy.

Polyp size quanti cation is currently the main variable for deciding the patient treatment during an endoscopic procedure. Nowadays, the polyp size is estimated by an expert, even when using devices that are provided with calibrated grids. As such estimation is highly subjective, automatic approaches have come to be appealing but also challenging because the polyp shape and appearance variability, the di erent types of motion present during the capture and the specular highlight noise. This work presents a novel approach to automatically estimate gastrointestinal polyp shapes in a video endoscopic sequence using spatiotemporal information. For doing so, a local spatio temporal descriptor is built up to obtain an initial segmentation since the polyp is the region with more movement. Then, an initial polyp manual segmentation outlines a region of interest (RoI) in the rst frame of the sequence and used as a reference for the polyp tracking during the sequence. Afterward, an exhaustive cross-correlation of the initial shape is carried out along the sequence and fused with the motion descriptor to re ne the original segmentation. The proposed approach was evaluated in 15 real video sequences achieving an average DSC score of 0:67% .

Accurate segmentation of the prostate and organs at risk is the fundamental guide for planning any radiotherapy. Such task is currently performed using a manual delineation of the organ on the MRI, a highly time consuming responsibility which in addition introduces inter and intra expert variability. Automatic MRI segmentation is a very challenging goal because of the large organ variability and the proximity of the neighboring organs. This work presents an automatic atlas-based segmentation strategy that selects the most probable template from a database using a robust multiscale similarity analysis. Once that probable template is selected, the associated segmentation is non-rigidly registered to the new MRI. The proposed method takes advantage of both the interindividual shape variation and intra-individual salient point representation. Results show that the method produces reliable segmentations, obtaining an average Dice Coefficient of 72% when comparing with the expert manual segmentation under a leave-one-out scheme with the training database.

This paper describes an extension of the Bag-of-Visual-Words (BoVW) representation for image categorization (IC) of histophatology images. This representation is one of the most used approaches in several high-level computer vision tasks. However, the BoVW representation has an important limitation: the disregarding of spatial information among visual words. This information may be useful to capture discriminative visual-patterns in specific computer vision tasks. In order to overcome this problem we propose the use of visual n-grams. N-grams based-representations are very popular in the field of natural language processing (NLP), in particular within text mining and information retrieval. We propose building a codebook of n-grams and then representing images by histograms of visual n-grams. We evaluate our proposal in the challenging task of classifying histopathology images. The novelty of our proposal lies in the fact that we use n-grams as attributes for a classification model (together with visual-words, i.e., 1-grams). This is common practice within NLP, although, to the best of our knowledge, this idea has not been explored yet within computer vision. We report experimental results in a database of histopathology images where our proposed method outperforms the traditional BoVWs formulation.

The advent of low-cost digital storage and automated microscope motor stages has made it possible to sequen- tially capture the full set of elds of view (FOV) covering a slide sample in a process commonly resulting in hundreds or even thousands of images. Aligning or verifying manually the alignment of thousands of images is an unrealistically labor intensive task that is, however, a fundamental step in the analysis of virtual-slides: For virtual-Slide creation the automation of the aligning process is not a mere an option but an absolute necessity. In the present work we propose the use of feature based methods and local consistency measures to improve the creation of mosaics from individual images captured with an in-house built microscope.

The aim of this study was to compare the diagnostic accuracy of the consumer-grade and medical-grade monitors —with very different costs— in breast cancer detection, when using with and without Digital Imaging and Communication in Medicine (DICOM) standard calibration. This was a retrospective study with factorial design and repeated measures, using 70 digital mammograms (40 benign or normal cases and 30 malignant cases), four radiologists, and three displays, with and without display calibration. Film mammograms were also included. Readings were classified according to the Breast Imaging Reporting and Data System. One medical-grade grayscale display and two consumer-grade displays were compared. Receiver operating characteristics curves were plotted for nodules, micro calcifications and the degree of malignancy. The diagnostic accuracy for each device was calculated as the area under these curves and accuracies were compared using analysis of variance.

There have existed for a long period several strategies developed by international organisms to improve their intervention at the very rst level of some public health problems. In particular, the prenatal control has been introduced as a structured strategy for the rst level as the integrated management of childhood illness (AIEPI in spanish) since more than twenty years. This paper presents a novel approach to include recent technological advances within the work ow of such process so that it facilitates interaction and decreases the training time. The method, named herein TeleAIEPI, implements the whole AIEPI questionnaire in a mobile application with high portability, little computational requirements and usability. The success of teleAIEPI application is completely dependent on the usability and integrability with any mobile device. The architecture, functional requirements and usability evaluation are herein presented, showing an adequate performance when real users interact with such an application.

Virtual microscopy (VM) facilitates visualization and deployment of histopathological virtual slides (VS), a useful tool for education, research and diagnosis. In recent years, it has become popular, yet its use is still limited basically because of the very large sizes of VS, typically of the order of gigabytes. Such volume of data requires efficacious and efficient strategies to access the VS content. In an educative or research scenario, several users may require to access and interact with VS at the same time, so, due to large data size, a very expensive and powerful infrastructure is usually required. This article introduces a novel JPEG2000-based service oriented architecture for streaming and visualizing very large images under scalable strategies, which in addition need not require very specialized infrastructure. Results suggest that the proposed architecture enables transmission and simultaneous visualization of large images, while it is efficient using resources and offering users proper response times.

Among non-invasive techniques, heart rate variability (HRV) analysis has become widely used for assessing the balance of the autonomic nervous system. Research in this area has not stopped and alternative tools for the study and interpretation of HRV, are still being proposed. Nevertheless, frequency-domain analysis of HRV is controversial when the heartbeat sequence is non-stationary. The Hilbert-Huang Transform (HHT) is a relative new technique for timefrequency analyses of non-linear and non-stationary signals. The main purpose of this work is to investigate the influence of time series´ length and noise in HRV from synthetic signals, using HHT and to compare it with Welch method. Synthetic heartbeat time series with different sizes and levels of signal to noise ratio (SNR) were investigated. Results shows i) sequence´s length did not affect the estimation of HRV spectral parameter, ii) favorable performance for HHT for different SNR. Additionally, HHT can be applied to non-stationary signals from nonlinear systems and it will be useful to HRV analysis to interpret autonomic activity when acute and transient phenomena are assessed.

This article presents a study of ventricular repolarization in diabetic and metabolic syndrome subjects. The corrected QT interval (QTc) was estimated using four correction formulas commonly employed in the literature: Bazett, Fridericia, Framingham and Hodges. After extracting the Q, R and T waves from the electrocardiogram of 52 subjects (19 diabetic, 15 with metabolic syndrome and 18 control), using a wavelet-based approach, the RR interval and QT interval were determined. Then, QTc interval was computed using the formulas previously mentioned. Additionally, laboratory test (fasting glucose, cholesterol, triglycerides) were also evaluated. Results show that metabolic syndrome subjects have normal QTc. However, a longer QTc in this population may be a sign of future complication. The corrected QT interval by Fridericia's formula seems to be the most appropriated for metabolic syndrome subjects (low correlation coefficient between RR and QTc). Significant differences were obtained in the blood glucose and triglyceride levels, principally due to the abnormal sugar metabolization of metabolic syndrome and diabetic subjects. Further studies are focused on the acquisition of a larger database of metabolic syndrome and diabetics subjects and the repetition of this study using other populations, like high performance athletes.

Current medical image processing has become a complex mixture of many scienti c disciplines including mathematics, statistics, physics, and algorithmics, to perform tasks such as registration, segmentation, and visualization, with the ultimate purpose of helping clinicians in their daily routine. This requires high performance computing capabilities that can be achieved in several ways, usually una ordable for most medical institutions. This paper presents a space-based computational grid that uses the otherwise wasted CPU cycles of a set of personal computers, to provide high-performance medical imaging services over the Internet. By using an existing hardware infrastructure and software of free distribution, the proposed approach is apt for university hospitals and other low-budget institutions. This will be illustrated by the use of three real case studies of services where an important speedup factor has been obtained and whose performance has become suitable for use in real clinical scenarios.

In this paper, we propose to use filtering methods and a segmentation algorithm for the analysis of fetal heart in ultrasound images. Since noise speckle makes difficult the analysis of ultrasound images, the filtering process becomes a useful task in these types of applications. The filtering techniques consider in this work assume that the speckle noise is a random variable with a Rayleigh distribution. We use two multiresolution methods: one based on wavelet decomposition and the another based on the Hermite transform. The filtering process is used as way to strengthen the performance of the segmentation tasks. For the wavelet-based approach, a Bayesian estimator at subband level for pixel classification is employed. The Hermite method computes a mask to find those pixels that are corrupted by speckle. On the other hand, we picked out a method based on a deformable model or "snake" to evaluate the influence of the filtering techniques in the segmentation task of left ventricle in fetal echocardiographic images.

Variational approaches based on level set representation have become some of the most important methodologies used to handle the segmentation tasks of biological structures in medical images. Because the segmentation is one of the most challenging processes in medical applications, all the methods fail to achieve perfect results. The major problems are due to noise, poor contrast and high variation of the structure shapes. In this paper, we review the principal level set – based methods that have been designed for image segmentation applications. These approaches include: Geodesic Active Contour, Chan-Vese Functional and Geodesic Active Regions. We also shortly analyze the first method proposed for shape extraction in images by using level set representation. We make a comparative study of the performance obtained for each method applied on cardiac CT images which present strong and very marked differences about the contrast and shape variation. Left ventricle is selected as structure of analysis. Measures of similarity are used to evaluate the performance of the methods.

This paper presents a shape based level set technique to extract vascular structures in X-Ray Angiographic images. It makes use of the Mumford-Shah functional¹⁹ applied to images of non-uniform illumination. The shape model is computed using string matching techniques from a preliminary hierarchical multiphase method. Its performance, using different metrics, has been evaluated on a set of three angiographic image sequences by comparison with manual delineation. A sensitivity of 91.31 ± 3.52% and a specificity of 94.28 ± 9.17% were found in the quantitative validation analysis.

Early tumor detection is key in reducing the number of breast cancer death and screening mammography is one of the most widely available and reliable method for early detection. However, it is difficult for the radiologist to process with the same attention each case, due the large amount of images to be read. Computer aided detection (CADe) systems improve tumor detection rate; but the current efficiency of these systems is not yet adequate and the correct interpretation of CADe outputs requires expert human intervention. Computer aided diagnosis systems (CADx) are being designed to improve cancer diagnosis accuracy, but they have not been efficiently applied in breast cancer. CADx efficiency can be enhanced by considering the natural mirror symmetry between the right and left breast. The objective of this work is to evaluate co-registration algorithms for the accurate alignment of the left to right breast for CADx enhancement. A set of mammograms were artificially altered to create a ground truth set to evaluate the registration efficiency of DEMONs , and SPLINE deformable registration algorithms. The registration accuracy was evaluated using mean square errors, mutual information and correlation. The results on the 132 images proved that the SPLINE deformable registration over-perform the DEMONS on mammography images.

Emphysema is associated with the destruction of lung parenchyma, resulting in abnormal enlargement of airspaces. Accurate quantification of emphysema is required for a better understanding of the disease as well as for the assessment of drugs and treatments. In the present study, a novel method for emphysema characterization from histological lung images is proposed. Elastase-induced mice were used to simulate the effect of emphysema on the lungs. A database composed of 50 normal and 50 emphysematous lung patches of size 512 x 512 pixels was used in our experiments. The purpose is to automatically identify those patches containing emphysematous tissue. The proposed approach is based on the use of granulometry analysis, which provides the pattern spectrum describing the distribution of airspaces in the lung region under evaluation. The profile of the spectrum was summarized by a set of statistical features. A logistic regression model was then used to estimate the probability for a patch to be emphysematous from this feature set. An accuracy of 87% was achieved by our method in the classification between normal and emphysematous samples. This result shows the utility of our granulometry-based method to quantify the lesions due to emphysema.

The main simulators of digestive surgery have been developed for solid organs such as the liver and spleen. Studies relating to soft tissues like the pelvic organs are rare. This is related to the kind of simulations in which the characterization of the physical parameters of the pelvic may present a complex task . In this paper, we consider two physical models, the rst one is elaborated using the mass-spring method and the second using the nite element method. Simulation results are presented and discussed on the basis of a comparison with a dynamic MRI data.

In this paper, we propose to take advantage from the contrast characteristics of our magnetic resonance images in order to improve the performance of Active Shape Models (ASM) applied to knee cartilage segmentation, since ASM depends directly of the contrast between objects. We realize an image fusion-based contrast enhancement between slices from magnetic resonance image volumes, then, we test the ASM algorithm with contrast enhancement images and compare results with ASM without contrast enhancement. The results are very clear, the ASM with contrast enhancement has a better performance and consistence, and we validate this results using Dice coefficient and Hausdorff distance. Moreover, we validate contrast enhancement to assure that really we are improving the contrast image.

Knee osteoarthritis (OA) is characterized by the morphological degeneration of cartilage. Efficient segmentation of cartilage is important for cartilage damage diagnosis and to support therapeutic responses. We present a method for knee cartilage segmentation in magnetic resonance images (MRI). Our method incorporates the Hermite Transform to obtain a hierarchical decomposition of contours which describe knee cartilage shapes. Then, we compute a statistical model of the contour of interest from a set of training images. Thereby, our Hierarchical Active Shape Model (HASM) captures a large range of shape variability even from a small group of training samples, improving segmentation accuracy. The method was trained with a training set of 16- MRI of knee and tested with leave-one-out method.