This PDF file contains the front matter associated with SPIE Proceedings Volume 9287, including the Title Page, Copyright information, Table of Contents, Author Index, and Conference Committee listing.

Nuclear architecture or the spatial arrangement of individual cancer nuclei on histopathology images has been shown to be associated with different grades and differential risk for a number of solid tumors such as breast, prostate, and oropharyngeal. Graph-based representations of individual nuclei (nuclei representing the graph nodes) allows for mining of quantitative metrics to describe tumor morphology. These graph features can be broadly categorized into global and local depending on the type of graph construction method. While a number of local graph (e.g. Cell Cluster Graphs) and global graph (e.g. Voronoi, Delaunay Triangulation, Minimum Spanning Tree) features have been shown to associated with cancer grade, risk, and outcome for different cancer types, the sensitivity of the preceding segmentation algorithms in identifying individual nuclei can have a significant bearing on the discriminability of the resultant features. This therefore begs the question as to which features while being discriminative of cancer grade and aggressiveness are also the most resilient to the segmentation errors. These properties are particularly desirable in the context of digital pathology images, where the method of slide preparation, staining, and type of nuclear segmentation algorithm employed can all dramatically affect the quality of the nuclear graphs and corresponding features. In this paper we evaluated the trade off between discriminability and stability of both global and local graph-based features in conjunction with a few different segmentation algorithms and in the context of two different histopathology image datasets of breast cancer from whole-slide images (WSI) and tissue microarrays (TMA). Specifically in this paper we investigate a few different performance measures including stability, discriminability and stability vs discriminability trade off, all of which are based on p-values from the Kruskal-Wallis one-way analysis of variance for local and global graph features. Apart from identifying the set of local and global features that satisfied the trade off between stability and discriminability, our most interesting finding was that a simple segmentation method was sufficient to identify the most discriminant features for invasive tumour detection in TMAs, whereas for tumour grading in WSI, the graph based features were more sensitive to the accuracy of the segmentation algorithm employed.

Computational anatomy is a subdiscipline of the anatomy that studies macroscopic details of the human body structure using a set of automatic techniques. Different reference systems have been developed for brain mapping and morphometry in functional and structural studies. Several models integrate particular anatomical regions to highlight pathological patterns in structural brain MRI, a really challenging task due to the complexity, variability, and nonlinearity of the human brain anatomy. In this paper, we present a strategy that aims to find anatomical regions with pathological meaning by using a probabilistic analysis. Our method starts by extracting visual primitives from brain MRI that are partitioned into small patches and which are then softly clustered, forming different regions not necessarily connected. Each of these regions is described by a co- occurrence histogram of visual features, upon which a probabilistic semantic analysis is used to find the underlying structure of the information, i.e., separated regions by their low level similarity. The proposed approach was tested with the OASIS data set which includes 69 Alzheimer’s disease (AD) patients and 65 healthy subjects (NC).

Morphometry based methods allow the detection of subtle anatomical differences in the Magnetic Resonance Images (MRI) between healthy subjects and Alzheimer's Disease (AD) patients. However, anatomical volumes are rarely used for clinical diagnosis as the changes induced by AD are hard to differentiate from normal brain aging.

We present a morphometry method which uses brain models generated using Nonnegative Matrix Factorization (NMF) characterized by signatures calculated from perceptual features such as intensities, edges or orientations, of salient regions. The Earth Mover's Distance (EMD), a robust measure of the cost of transforming signature A into signature B, is used to calculate volume-models distances. The discerning power of these distances is tested by using them as features for a Support Vector Machine classifier.

This work shows the usefulness of the EMD as a metric in medical image applications as it has proven to be robust to bin selection, takes into account cross bin relations, and allows high sensitivity with lower dimensionality. This method is able to find discerning regions which, besides aiding in classification, may provide new insights of the disease's development.

We develop a new algorithm to compute voxel-wise shape differences in tensor-based morphometry (TBM). As in standard TBM, we non-linearly register brain T1-weighed MRI data from a patient and control group to a template, and compute the Jacobian of the deformation fields. In standard TBM, the determinants of the Jacobian matrix at each voxel are statistically compared between the two groups. More recently, a multivariate extension of the statistical analysis involving the deformation tensors derived from the Jacobian matrices has been shown to improve statistical detection power.⁷ However, multivariate methods comprising large numbers of variables are computationally intensive and may be subject to noise. In addition, the anatomical interpretation of results is sometimes difficult. Here instead, we analyze the eigenvalues and the eigenvectors of the Jacobian matrices. Our method is validated on brain MRI data from Alzheimer’s patients and healthy elderly controls from the Alzheimer’s Disease Neuro Imaging Database.

Fetal Magnetic Resonance (FMR) is an imaging technique that is becoming increasingly important as allows assessing brain development and thus make an early diagnostic of congenital abnormalities, spatial resolution is limited by the short acquisition time and the unpredictable fetus movements, in consequence the resulting images are characterized by non-parallel projection planes composed by anisotropic voxels. The sparse Bayesian representation is a flexible strategy which is able to model complex relationships. The Super-resolution is approached as a regression problem, the main advantage is the capability to learn data relations from observations. Quantitative performance evaluation was carried out using synthetic images, the proposed method demonstrates a better reconstruction quality compared with standard interpolation approach. The presented method is a promising approach to improve the information quality related with the 3-D fetal brain structure. It is important because allows assessing brain development and thus make an early diagnostic of congenital abnormalities.

This work addresses the problem of lung sound classification, in particular, the problem of distinguishing between wheeze and normal sounds. Wheezing sound detection is an important step to associate lung sounds with an abnormal state of the respiratory system, usually associated with tuberculosis or another chronic obstructive pulmonary diseases (COPD). The paper presents an approach for automatic lung sound classification, which uses different state-of-the-art sound features in combination with a C-weighted support vector machine (SVM) classifier that works better for unbalanced data. Feature extraction methods used here are commonly applied in speech recognition and related problems thanks to the fact that they capture the most informative spectral content from the original signals. The evaluated methods were: Fourier transform (FT), wavelet decomposition using Wavelet Packet Transform bank of filters (WPT) and Mel Frequency Cepstral Coefficients (MFCC). For comparison, we evaluated and contrasted the proposed approach against previous works using different combination of features and/or classifiers. The different methods were evaluated on a set of lung sounds including normal and wheezing sounds. A leave-two-out per-case cross-validation approach was used, which, in each fold, chooses as validation set a couple of cases, one including normal sounds and the other including wheezing sounds. Experimental results were reported in terms of traditional classification performance measures: sensitivity, specificity and balanced accuracy. Our best results using the suggested approach, C-weighted SVM and MFCC, achieve a 82.1% of balanced accuracy obtaining the best result for this problem until now. These results suggest that supervised classifiers based on kernel methods are able to learn better models for this challenging classification problem even using the same feature extraction methods.

Heartbeat characterization is an important issue in cardiac assistance diagnosis systems. In particular, wide sets of features are commonly used in long term electrocardiographic signals. Then, if such a feature space does not represent properly the arrhythmias to be grouped, classification or clustering process may fail. In this work a suitable feature set for different heartbeat types is studied, involving morphology, representation and time-frequency features. To determine what kind of features generate better clusters, feature selection procedure is used and assessed by means clustering validity measures. Then the feature subset is shown to produce fine clustering that yields into high sensitivity and specificity values for a broad range of heartbeat types.

Physiological signals are commonly the result of complex interactions between systems and organs, these interactions lead to signals that exhibit a non-stationary behaviour. For cardiac signals, non-stationary heart rate variability (HRV) may produce misinterpretations. A previous work proposed to divide a non-stationary signal into stationary segments by looking for changes in the signal’s properties related to changes in the mean of the signal. In this paper, we extract stationary segments from non-stationary synthetic and cardiac signals. For synthetic signals with different signal-to-noise ratio levels, we detect the beginning and end of the stationary segments and the result is compared to the known values of the occurrence of these events. For cardiac signals, RR interval (cardiac cycle length) time series, obtained from electrocardiographic records during stress tests for two populations (diabetic patients with cardiovascular autonomic neuropathy and control subjects), were divided into stationary segments. Results on synthetic signals reveal that the non-stationary sequence is divided into more stationary segments than needed. Additionally, due to HRV reduction and exercise intolerance reported on diabetic cardiovascular autonomic neuropathy patients, non-stationary RR interval sequences from these subjects can be divided into longer stationary segments compared to the control group.

Using multiparametric MRI (mpMRI) protocols to monitor prostate cancer could provide new insights into the biological mechanisms of developing tumours. Automatically discriminating tumour regions active area of research due to the complexity and plurality of cancer behaviour. This work evaluates four different Magnetic Resonance Imaging (MRI) image modalities, namely: Diffusion-Weighted Imaging evaluated at b = {0, 100, 1000}, Apparent Diffusion Coefficient and Dynamic Contrast Enhanced MRI, by extracting texture and functional features and then selecting the optimal ones to discriminate anatomical prostate regions in each modality. The images used were taken prior to radiotherapy from eight patients previously diagnosed with moderate risk of recurrent cancer. Finally, we compared the relevance of each modality to discriminate between healthy tissue and tumour cells.

The pelvic magnetic Resonance images (MRI) are used in Prostate cancer radiotherapy (RT), a process which is part of the radiation planning. Modern protocols require a manual delineation, a tedious and variable activity that may take about 20 minutes per patient, even for trained experts. That considerable time is an important work ow burden in most radiological services. Automatic or semi-automatic methods might improve the efficiency by decreasing the measure times while conserving the required accuracy. This work presents a fully automatic atlas- based segmentation strategy that selects the more similar templates for a new MRI using a robust multi-scale SURF analysis. Then a new segmentation is achieved by a linear combination of the selected templates, which are previously non-rigidly registered towards the new image. The proposed method shows reliable segmentations, obtaining an average DICE Coefficient of 79%, when comparing with the expert manual segmentation, under a leave-one-out scheme with the training database.

Breast parenchymal density is considered a strong indicator of cancer risk. However, measures of breast density are often qualitative and require the subjective judgment of radiologists. This work proposes a supervised algorithm to automatically assign a BI-RADS breast density score to a digital mammogram. The algorithm applies principal component analysis to the histograms of a training dataset of digital mammograms to create four different spaces, one for each BI-RADS category. Scoring is achieved by projecting the histogram of the image to be classified onto the four spaces and assigning it to the closest class. In order to validate the algorithm, a training set of 86 images and a separate testing database of 964 images were built. All mammograms were acquired in the craniocaudal view from female patients without any visible pathology. Eight experienced radiologists categorized the mammograms according to a BIRADS score and the mode of their evaluations was considered as ground truth. Results show better agreement between the algorithm and ground truth for the training set (kappa=0.74) than for the test set (kappa=0.44) which suggests the method may be used for BI-RADS classification but a better training is required.

In prostate cancer external beam radiotherapy, pelvic structures identification in computed tomography (CT) is required for the treatment planning and is performed manually by experts. Prostate manual delineations in CT modality is time consuming and prone to observer variability. We propose a fully automated process using a combination of a Random Forests (RF) classification and Spherical Harmonics (SPHARM) to identify the prostate boundaries. The proposed method outperformed classical atlas based approach from the literature. Combining RF to detect the prostate and SPHARM for shape regularization provided promising results for automatic prostate segmentation.

Learning data representations directly from the data itself is an approach that has shown great success in different pattern recognition problems, outperforming state-of-the-art feature extraction schemes for different tasks in computer vision, speech recognition and natural language processing. Representation learning applies unsupervised and supervised machine learning methods to large amounts of data to find building-blocks that better represent the information in it. Digitized histopathology images represents a very good testbed for representation learning since it involves large amounts of high complex, visual data. This paper presents a comparative evaluation of different supervised and unsupervised representation learning architectures to specifically address open questions on what type of learning architectures (deep or shallow), type of learning (unsupervised or supervised) is optimal. In this paper we limit ourselves to addressing these questions in the context of distinguishing between anaplastic and non-anaplastic medulloblastomas from routine haematoxylin and eosin stained images. The unsupervised approaches evaluated were sparse autoencoders and topographic reconstruct independent component analysis, and the supervised approach was convolutional neural networks. Experimental results show that shallow architectures with more neurons are better than deeper architectures without taking into account local space invariances and that topographic constraints provide useful invariant features in scale and rotations for efficient tumor differentiation.

Accessing information of interest in collections of histopathology images is a challenging task. To address such issue, previous works have designed searching strategies based on the use of keywords and low-level features. However, those methods have demonstrated to not be enough or practical for this purpose. Alternative low-level features such as cell area, distance among cells and cell density are directly associated to simple histological concepts and could serve as good descriptors for this purpose. In this paper, a statistical model is adapted to represent the distribution of the areas occupied by cells for its use in whole histopathology image characterization. This novel descriptor facilitates the design of metrics based on distribution parameters and also provides new elements for a better image understanding. The proposed model was validated using image processing and statistical techniques. Results showed low error rates, demonstrating the accuracy of the model.

Conditions such as surgical procedures or vascular diseases produce arterial ischemia and reperfusion injuries, which generate changes in peripheral tissues and organs, for instance, in striated skeletal muscle. To determine such changes, we conducted an experimental method in which 42 male Wistar rat were selected, to be undergone to tourniquet application on the right forelimb and left hind limb, to induce ischemia during one and three hours, followed by reperfusion periods starting at one hour and it was prolonged up to 32 days. Extensor carpi radialis longus and soleus respectively, were obtained to be processed for histochemical and morphometric analysis. By means of image processing and detection of regions of interest, variations of areas occupied by muscle fibers and intramuscular extracellular matrix (IM-ECM) throughout reperfusion were observed. In extensor carpi radialis longus, results shown reduction in the area occupied by muscle fibers; this change is significant between one hour and three hours ischemia followed by 16 hours, 48 hours and 32 days reperfusión (p˂0.005). To compare only periods of reperfusión that continued to three hours ischemia, were found significant differences, as well. For area occupied by IM-ECM, were identified increments in extensor carpi radialis longus by three hours ischemia and eight to 16 days reperfusion; in soleus, was observed difference by one hour ischemia with 42 hours reperfusion, and three hours ischemia followed by four days reperfusion (p˂0.005). Skeletal muscle develops adaptive changes in longer reperfusion, to deal with induced injury. Descriptions beyond 32 days reperfusion, can determine recovering normal pattern.

Telecytology is a new research area that holds the potential of significantly reducing the number of deaths due to cervical cancer in developing countries. This work presents a novel super-resolution technique that couples high and low frequency information in order to reduce the bandwidth consumption of cervical image transmission. The proposed approach starts by decomposing into wavelets the high resolution images and transmitting only the lower frequency coefficients. The transmitted coefficients are used to reconstruct an image of the original size. Additional details are added by iteratively replacing patches of the wavelet reconstructed image with equivalent high resolution patches from a previously acquired image database. Finally, the original transmitted low frequency coefficients are used to correct the final image. Results show a higher signal to noise ratio in the proposed method over simply discarding high frequency wavelet coefficients or replacing directly down-sampled patches from the image-database.

The left ventricle (LV) segmentation plays an important role in a subsequent process for the functional analysis of the LV. Typical segmentation of the endocardium wall in the ventricle excludes papillary muscles which leads to an incorrect measure of the ejected volume in the LV. In this paper we present a new variational strategy using a 2D level set framework that includes a local term for enhancing the low contrast structures and a 2D shape model. The shape model in the level set method is propagated to all image sequences corresponding to the cardiac cycles through the optical flow approach using the Hermite transform. To evaluate our strategy we use the Dice index and the Hausdorff distance to compare the segmentation results with the manual segmentation carried out by the physician.

Auscultation is one of the most utilized physical examination procedures for listening to lung, heart and intestinal sounds during routine consults and emergencies. Heart and lung sounds overlap in the thorax. An algorithm was used to separate them based on the discrete wavelet transform with multi-resolution analysis, which decomposes the signal into approximations and details. The algorithm was implemented in software and in hardware to achieve real-time signal separation. The heart signal was found in detail eight and the lung signal in approximation six. The hardware was used to separate the signals with a delay of 256 ms. Sending wavelet decomposition data - instead of the separated full signa - allows telemedicine applications to function in real time over low-bandwidth communication channels.

Cardiac arrhythmia analysis on Holter recordings is an important issue in clinical settings, however such issue implicitly involves attending other problems related to the large amount of unlabelled data which means a high computational cost. In this work an unsupervised methodology based in a segment framework is presented, which consists of dividing the raw data into a balanced number of segments in order to identify fiducial points, characterize and cluster the heartbeats in each segment separately. The resulting clusters are merged or split according to an assumed criterion of homogeneity. This framework compensates the high computational cost employed in Holter analysis, being possible its implementation for further real time applications. The performance of the method is measure over the records from the MIT/BIH arrhythmia database and achieves high values of sensibility and specificity, taking advantage of database labels, for a broad kind of heartbeats types recommended by the AAMI.

In this paper, we propose an automated method to classify normal/abnormal wall motion in Left Ventricle (LV) function in cardiac cine-Magnetic Resonance Imaging (MRI), taking as reference, strain information obtained from 2D Speckle Tracking Echocardiography (STE). Without the need of pre-processing and by exploiting all the images acquired during a cardiac cycle, spatio-temporal profiles are extracted from a subset of radial lines from the ventricle centroid to points outside the epicardial border. Classical Support Vector Machines (SVM) are used to classify features extracted from gray levels of the spatio-temporal profile as well as their representations in the Wavelet domain under the assumption that the data may be sparse in that domain. Based on information obtained from radial strain curves in 2D-STE studies, we label all the spatio-temporal profiles that belong to a particular segment as normal if the peak systolic radial strain curve of this segment presents normal kinesis, or abnormal if the peak systolic radial strain curve presents hypokinesis or akinesis. For this study, short-axis cine- MR images are collected from 9 patients with cardiac dyssynchrony for which we have the radial strain tracings at the mid-papilary muscle obtained by 2D STE; and from one control group formed by 9 healthy subjects. The best classification performance is obtained with the gray level information of the spatio-temporal profiles using a RBF kernel with 91.88% of accuracy, 92.75% of sensitivity and 91.52% of specificity.

Sports related traumatic brain injury (TBI) is a worldwide public health issue, and damage to the corpus callosum (CC) has been considered as an important indicator of TBI. However, contact sports players suffer repeated hits to the head during the course of a season even in the absence of diagnosed concussion, and less is known about their effect on callosal anatomy. In addition, T1-weighted and diffusion tensor brain magnetic resonance images (DTI) have been analyzed separately, but a joint analysis of both types of data may increase statistical power and give a more complete understanding of anatomical correlates of subclinical concussions in these athletes. Here, for the first time, we fuse T1 surface-based morphometry and a new DTI analysis on 3D surface representations of the CCs into a single statistical analysis on these subjects. Our new combined method successfully increases detection power in detecting differences between pre- vs. post-season contact sports players. Alterations are found in the ventral genu, isthmus, and splenium of CC. Our findings may inform future health assessments in contact sports players. The new method here is also the first truly multimodal diffusion and T1-weighted analysis of the CC, and may be useful to detect anatomical changes in the corpus callosum in other multimodal datasets.

Investigations about the intrinsic brain organization in resting-state are critical for the understanding of healthy, pathological and pharmacological cerebral states. Recent studies on fMRI suggest that resting state activity is organized on large scale networks of coordinated activity, in the so called, Resting State Networks (RSNs). The assessment of the interactions among these functional networks plays an important role for the understanding of different brain pathologies. Current methods to quantify these interactions commonly assume that the underlying coordination mechanisms are stationary and linear through the whole recording of the resting state phenomena. Nevertheless, recent evidence suggests that rather than stationary, these mechanisms may exhibit a rich set of time-varying repertoires. In addition, these approaches do not consider possible non-linear relationships maybe linked to feed-back communication mechanisms between RSNs. In this work, we introduce a novel approach for dynamical functional network connectivity for functional magnetic resonance imaging (fMRI) resting activity, which accounts for non-linear dynamic relationships between RSNs. The proposed method is based on a windowed distance correlations computed on resting state time-courses extracted at single subject level. We showed that this strategy is complementary to the current approaches for dynamic functional connectivity and will help to enhance the discrimination capacity of patients with disorder of consciousness.

Disorders of consciousness (DOC) are a consequence of a variety of severe brain injuries. DOC commonly results in anatomical brain modifications, which can affect cortical and sub-cortical brain structures. Postmortem studies suggest that severity of brain damage correlates with level of impairment in DOC. In-vivo studies in neuroimaging mainly focus in alterations on single structures. Recent evidence suggests that rather than one, multiple brain regions can be simultaneously affected by this condition. In other words, DOC may be linked to an underlying cerebral network of structural damage. Recently, geometrical spatial relationships among key sub-cortical brain regions, such as left and right thalamus and brain stem, have been used for the characterization of this network. This approach is strongly supported on automatic segmentation processes, which aim to extract regions of interests without human intervention. Nevertheless, patients with DOC usually present massive structural brain changes. Therefore, segmentation methods may highly influence the characterization of the underlying cerebral network structure. In this work, we evaluate the level of characterization obtained by using the spatial relationships as descriptor of a sub-cortical cerebral network (left and right thalamus) in patients with DOC, when different segmentation approaches are used (FSL, Free-surfer and manual segmentation). Our results suggest that segmentation process may play a critical role for the construction of robust and reliable structural characterization of DOC conditions.

Disorders of consciousness (DOC) may be characterized by the degree at which consciousness is impaired, and include for example vegetative state (VS) and minimally conscious state (MCS) patients. Using a reliable marker as a measure of the level of consciousness in such patients is of utmost necessity and importance for their appropriate diagnosis and prognosis. Identification of VS and MCS states based on their behaviors sometimes leads to incorrect inferences due to the influence of a range of factors like motor impairment, fluctuating arousal levels and rapidly habituating responses to name a few.¹ The extent of damage in the thalamus, a structure known for its role in arousal regulation, may provide an imaging biomarker to better differentiate between VS and MCS. In this study, we manually segmented the thalamus from T1-weighted brain MRI images in a large cohort of 19 VS and 23 MCS subjects that were examined using the French version of the Coma Recovery Scale Revised (CRS-R).² This scale is the most trustworthy behavioural diagnosis tool³ for patients with DOC available. The aim was to determine whether a relationship between thalamus volume and consciousness level exists. Results show that total thalamic volume tends to decrease over time after a severe brain injury. Moreover, for subjects in chronic state, the thalamic volume seems to differ with respect to the degree of consciousness that was diagnosed. Finally, for these same chronic patients, the total thalamic volume is varying linearly as a function of the CRS-R score obtained, indicating that thalamic volume may be used as a biomarker to measure the level of consciousness.

Protontherapy is based on physical properties of ion beams which allow the delivery of high radiation doses at very precise location in the body of the patient. The treatment planning aims at maximizing the delivery in the target volume while avoiding any organs at risk. The treatment is generally planned prior the treatment, and the patient is aligned in the treatment room on the basis of fiducial markers. However, the alignment of the patient may suffer from lack of precision and moreover, the body of the patient may vary between the time of imaging for planning and the time of treatment in the protontherapy room. More precise protontherapy and adaptive treatment which can track modifications of the body and the treatment of mobile tumors require the design of in vivo imaging systems to be deployed in the treatment room. The goal of this paper is to overview the present and future development of in-vivo image guided protontherapy and to give some image processing related challenges. The technique mostly used today is to take 2 orthogonal X-ray views of the patient. It requires an efficient 2D-3D coregistration procedure but is quite easy to deploy. Cone Beam CT is a next step which allows the capture of an in-vivo 3-D view on which the 3-D planning can be registered. The ultimate goal is to develop 4-D imaging techniques suited for the treatment of mobile tumors, for the cases of lung cancer. The development of new detectors will allow to validate the treatment by an “a posteriori” validation of the dose delivery in the body.

Image registration aims to identify the mapping between corresponding locations in an anatomic structure. Most traditional approaches solve this problem by minimizing some error metric. However, they do not quantify the uncertainty behind their estimates and the feasibility of other solutions. In this work, it is assumed that two images of the same anatomic structure are related via a Lipschitz non-rigid deformation (the registration map). An approach for identifying point correspondences with zero false-negative rate and high precision is introduced under this assumption. This methodology is then extended to registration of regions in an image which is posed as a graph matching problem with geometric constraints. The outcome of this approach is a homeomorphism with uncertainty bounds characterizing its accuracy over the entire image domain. The method is tested by applying deformation maps to the LPBA40 dataset.

Reconstruction of the heartbeat is an useful tool to detect and diagnose some pathologies. However, this process represents a challenge because the heart is a moving organ inside a moving body, so that, either similar regions are hard to identify or some regions appear and disappear constantly. This article presents a reconstruction method of the right ventricle using SURF points in irregular regions. The SURF points, invariant to image scale and rotation, provide robust features of a right ventricle slice that can then be traced to the other slices. By using such points and then, using a nonrigid registration, it possible to perform a volumetrical reconstruction of these images.

Banks of high-quality, multimodal neurological images offer new possibilities for analyses based on brain registration. To take full advantage of these, current algorithms should be significantly enhanced. We present here a new brain registration method driven simultaneously by the structural intensity and the total diffusion information of MRI scans. Using the two modalities together allows for a better alignment of general and specific aspects of the anatomy. Furthermore, keeping the full diffusion tensor in the cost function, rather than only some of its scalar measures, will allow for a thorough statistical analysis once the Jacobian of the transformation is obtained.

Computer aided diagnosis systems (CAD) have been developed to assist radiologists in the detection and diagnosis of abnormalities and a large number of pattern recognition techniques have been proposed to obtain a second opinion. Most of these strategies have been evaluated using different datasets making their performance incomparable. In this work, an open access database of thyroid ultrasound images is presented. The dataset consists of a set of B-mode Ultrasound images, including a complete annotation and diagnostic description of suspicious thyroid lesions by expert radiologists. Several types of lesions as thyroiditis, cystic nodules, adenomas and thyroid cancers were included while an accurate lesion delineation is provided in XML format. The diagnostic description of malignant lesions was confirmed by biopsy. The proposed new database is expected to be a resource for the community to assess different CAD systems.

The monitoring of cyclists during physical activity is an important factor to improve their performance. We discuss a new approaches based on smartphone for monitoring physiological signal wirelessly for cyclists, using a pulse oximeter sensor attached to the rider’s forehead. This paper presents a wireless pulse Oximeter that was developed with a Nellcor’s module, which uses the Standard Host Interface Protocol (SHIP) for communication with the Bluetooth module and sends data for a Smartphone with Android O.S. Then these data are shown in the screen: the heartbeat and saturation percentage. The application was created with App Inventor and the data are sent to Google Maps via Twitter. The results demonstrate the possibility of developing a successful prototype.

Virtual rehabilitation (VR) is a novel motor rehabilitation therapy in which the rehabilitation exercises occurs through interaction with bespoken virtual environments. These virtual environments dynamically adapt their activity to match the therapy progress. Adaptation should be guided by the cognitive and emotional state of the patient, none of which are directly observable. Here, we present our first steps towards inferring non-observable attentional state from unobtrusively observable seated posture, so that this knowledge can later be exploited by a VR platform to modulate its behaviour. The space of seated postures was discretized and 648 pictures of acted representations were exposed to crowd-evaluation to determine attributed state of attention. A semi-supervised classifier based on Na¨ıve Bayes with structural improvement was learnt to unfold a predictive relation between posture and attributed attention. Internal validity was established following a 2×5 cross-fold strategy. Following 4959 votes from crowd, classification accuracy reached a promissory 96.29% (µ±σ = 87.59±6.59) and F-measure reached 82.35% (µ ± σ = 69.72 ± 10.50). With the afforded rate of classification, we believe it is safe to claim posture as a reliable proxy for attributed attentional state. It follows that unobtrusively monitoring posture can be exploited for guiding an intelligent adaptation in a virtual rehabilitation platform. This study further helps to identify critical aspects of posture permitting inference of attention.

Communication is one of the essential needs of human beings. Augmentative and Alternative Communication Systems (AAC) seek to help in the generation of oral and written language to people with physical disorders that limit their natural communication. These systems present significant challenges such as: the composition of consistent messages according to syntactic and semantic rules, the improvement of message production times, the application to social contexts and, consequently, the incorporation of user-specific information. This work presents an original ontology personalization approach for an AAC instant messaging system incorporating personalized information to improve the efficacy and efficiency of the message production. This proposal is based on a projection of a general ontology into a more specific one, avoiding storage redundancy and data coupling, representing a big opportunity to enrich communication capabilities of current AAC systems. The evaluation was performed for a study case based on an AAC system for assistance in composing messages. The results show that adding user-specific information allows generation of enriched phrases, so improving the accuracy of the message, facilitating the communication process.

Tensor-Based Morphometry (TBM) is an increasingly popular method for group analysis of brain MRI data. The main steps in the analysis consist of a nonlinear registration to align each individual scan to a common space, and a subsequent statistical analysis to determine morphometric differences, or difference in fiber structure between groups. Recently, we implemented the Statistically-Assisted Fluid Registration Algorithm or SAFIRA,¹ which is designed for tracking morphometric differences among populations. To this end, SAFIRA allows the inclusion of statistical priors extracted from the populations being studied as regularizers in the registration. This flexibility and degree of sophistication limit the tool to expert use, even more so considering that SAFIRA was initially implemented in command line mode. Here, we introduce a new, intuitive, easy to use, Matlab-based graphical user interface for SAFIRA’s multivariate TBM. The interface also generates different choices for the TBM statistics, including both the traditional univariate statistics on the Jacobian matrix, and comparison of the full deformation tensors.² This software will be freely disseminated to the neuroimaging research community.

Chronic wounds are a major problem worldwide which mainly affects to the geriatric population or patients with limited mobility. In tropical countries, Cutaneous Leishmaniasis(CL)s is also a cause for chronic wounds,being endemic in Peru in the 75% of the country. Therefore, the monitoring of these wounds represents a big challenge due to the remote location of the patients. This papers aims to develop a low-cost user-friendly technique to obtain a 3D reconstruction for chronic wounds oriented to clinical monitoring and assessment. The video is taken using a commercial hand-held video camera without the need of a rig. The algorithm has been specially designed for skin wounds which have certain characteristics in texture where techniques used in regular SFM applications with undefined edges wouldn’t work. In addition, the technique has been developed using open source libraries. The 3D cloud point estimated allows the computation of metrics as volume, depth, superficial area which recently have been used by CL specialists showing good results in clinical assessment. Initial results in cork phantoms and CL wounds show an average distance error of less than 1mm when compared against models obtained with a industrial 3D laser scanner.

SemVisM is a toolbox that combines medical informatics and computer graphics tools for reducing the semantic gap between low-level features and high-level semantic concepts/terms in the images. This paper presents a novel strategy for visualizing medical data annotated semantically, combining rendering techniques, and segmentation algorithms. SemVisM comprises two main components: i) AMORE (A Modest vOlume REgister) to handle input data (RAW, DAT or DICOM) and to initially annotate the images using terms defined on medical ontologies (e.g., MesH, FMA or RadLex), and ii) VOLPROB (VOlume PRObability Builder) for generating the annotated volumetric data containing the classified voxels that belong to a particular tissue. SemVisM is built on top of the semantic visualizer ANISE.¹

Diffusion MRI allows us not only voxelized diffusion characteristics but also the potential to delineate neuronal fiber path through tractography. There is a dearth of flexible open source tractography software programs for visualizing these complicated 3D structures. Moreover, rendering these structures using various shading, lighting, and representations will result in vastly different graphical feel. In addition, the ability to output these objects in various formats increases the utility of this platform. We have created TractRender that leverages openGL features through Matlab, allowing for maximum ease of use but still maintain the flexibility of custom scene rendering.

We report a new method for adjusting the points of an active shape model (ASM) to the edge of an object, on a grey level image. The method is based on the original iterative search for an optimum location of each point of the ASM, along the normal direction to the model boundary. In this work we determine the optimum location of the model boundary point through minimization of the error (euclidean distance) between a profile of pixels sampled along the normal direction, and its projection on the principal component axes, obtained from a training set of normal pixel profiles, located at the edge of the object. We validated our method on ultrasound images of the prostate and photographs of the left hand. Significant improvements were observed in the segmentation of the ultrasound images, with reference to the original ASM method of adjustment, while no significant improvement was observed for the photographs. Our method produced a mean error of 4.58 (mm) between corresponding expert and automatically annotated contours of the ultrasound images of the prostate, and 3.12 (mm) of mean error for the photographs of the left hand.

The purpose of this study is to determine the effectiveness of segmentation of axial MR proton density (PD) images of bony humeral head. PD sequence images which are included in standard shoulder MRI protocol are used instead of T1 MR images. Bony structures were reported to be successfully segmented in the literature from T1 MR images. T1 MR images give more sharp determination of bone and soft tissue border but cannot address the pathological process which takes place in the bone. In the clinical settings PD images of shoulder are used to investigate soft tissue alterations which can cause shoulder instability and are better in demonstrating edema and the pathology but have a higher noise ratio than other modalities. Moreover the alteration of humeral head intensity in patients and soft tissues in contact with the humeral head which have the very similar intensities with bone makes the humeral head segmentation a challenging problem in PD images. However segmentation of the bony humeral head is required initially to facilitate the segmentation of the soft tissues of shoulder. In this study shoulder MRI of 33 randomly selected patients were included. Speckle reducing anisotropic diffusion (SRAD) method was used to decrease noise and then Active Contour Without Edge (ACWE) and Signed Pressure Force (SPF) models were applied on our data set. Success of these methods is determined by comparing our results with manually segmented images by an expert. Applications of these methods on PD images provide highly successful results for segmentation of bony humeral head. This is the first study to determine bone contours in PD images in literature.

In 2010, cardiovascular disease (CVD) caused 33% of the total deaths in Brazil. Modalities such as Intravascular Optical Coherent Tomography (IOCT) provides coronary in vivo for detecting and monitoring the progression of CVDs. Specifically, this type of modality is widely used in neo-intima post stent re-stenosis investigation. Computational methods applied to IOCT images can render objective structure information, such as areas, perimeters, etc., allowing more accurate diagnostics. However, the variety of methods in the literature applied in IOCT is still small compared to other related modalities. Therefore, we propose a stent segmentation approach based on extracted features by gradient operations, and Mathematical Morphology. The methodology can be summarized as following: the lumen is segmented and the contrast stretching is generated, both to be used as auxiliary information. Second, the edges of objects were obtained by gradient computation. Next, a stent extractor finds and select relevant stent information. Finally, an interpolation procedure followed by morphological operations ends the segmentation. To evaluate the method, 160 images from pig coronaries were segmented and compared to their gold standards, the images were acquired after 30, 90 and 180 days of stent implantation. The proposed approach present good accuracy of True Positive (TP(%)) = 96.51±5.10, False Positive (FP(%)) = 6.09±5.32 , False Negative (FN(%)) = 3.49±5.10. Conclusion, the good results and the low complexity encourage the use and continuous evolution of current approach. However, only images of IOCT-TD technology were evaluated; therefore, further investigations should adapt this approach to work with IOCT-FD technology as well.

Medical image watermarking is an open area for research and is a solution for the protection of copyright and intellectual property. One of the main challenges of this problem is that the marked images should not differ perceptually from the original images allowing a correct diagnosis and authentication. Furthermore, we also aim at obtaining watermarked images with very little numerical distortion so that computer vision tasks such as segmentation of important anatomical structures do not be impaired or affected. We propose a preliminary watermarking application in cardiac CT images based on a perceptive approach that includes a brightness model to generate a perceptive mask and identify the image regions where the watermark detection becomes a difficult task for the human eye. We propose a normalization scheme of the image in order to improve robustness against geometric attacks. We follow a spread spectrum technique to insert an alphanumeric code, such as patient’s information, within the watermark. The watermark scheme is based on the Hermite transform as a bio-inspired image representation model. In order to evaluate the numerical integrity of the image data after watermarking, we perform a segmentation task based on deformable models. The segmentation technique is based on a vector-value level sets method such that, given a curve in a specific image, and subject to some constraints, the curve can evolve in order to detect objects. In order to stimulate the curve evolution we introduce simultaneously some image features like the gray level and the steered Hermite coefficients as texture descriptors. Segmentation performance was assessed by means of the Dice index and the Hausdorff distance. We tested different mark sizes and different insertion schemes on images that were later segmented either automatic or manual by physicians.

A first diagnosis of colorectal cancer is performed by examination of polyp shape and appearance during an endoscopy routine procedure. However, the video-endoscopy is highly noisy because exacerbated physiological conditions like increased motility or secretion may limit the visual analysis of lesions. In this work a 3D reconstruction of the digestive tract is proposed, facilitating the polyp shape evaluation by highlighting its surface geometry and allowing an analysis from different perspectives. The method starts by a spatio-temporal map, constructed to group the different regions of the tract by their similar dynamic patterns during the sequence. Then, such map was convolved with a second derivative of a Gaussian kernel that emulates the camera distortion and allows to highlight the polyp surface. The position initialization in each frame of the kernel was computed from expert manual delineation and propagated along the sequence based on. Results show reliable reconstructions, with a salient 3D polyp structure that can then be better observed.

Traditionally, the Parkinson disease is diagnosed and followed up by conventional clinical tests that are fully dependent on the expert experience. The diffuse boundary between normal and early Parkinson stages and the high variability of gait patterns difficult any objective characterization of this disease. An automatic characterization of the disease is herein proposed by mixing up different measures of the ipsilateral coordination and spatiotemporal gait patterns which are then classified with a classical support vector machine. The strategy was evaluated in a population with Parkinson and healthy control subjects, obtaining an average accuracy of 87% for the task of classification.

Parkinson’s Disease characterization is commonly carried out by measuring a motor abnormality that may affect an optimal locomotion. However, such gait characterization is far from achieving accurate and sensible early detection of this disease, dealying between 6 months to 3 years a first diagnosis. Current research has identified the eye movements (EM) as a powerful biomarker that may detect and identify PD, even in early stages. However, this eye analysis is now performed under fully controlled conditions and strict protocols, for which the patient must follow a set of routine movements in a static position. Such protocols however loss some natural eye movements during the gait that may help to promptly highlight the disease. This work presents preliminary results characterizing and analyzing the center of mass of the eye movement during the gait, captured using a high speed camera. An automatic tracking strategy was herein implemented to follow the eye during the locomotion. Promising results were obtained from a set of real patients diagnosed with parkinson diseases in stages of 1 y 3, which show strong differences among the computed signals.

Parkinson’s disease is a neurodegenerative disorder that progressively affects the movement. Gait analysis is therefore crucial to determine a disease degree as well as to orient the diagnosis. However, gait examination is completely subjective and therefore prone to errors or misinterpretations, even with a great expertise. In addition, the conventional evaluation follows up general gait variables, which amounts to ignore subtle changes that definitely can modify the history of the treatment. This work presents a functional gait model that simulates the center of gravity trajectory (CoG) for different Parkinson disease stages. This model mimics the gait trajectory by coupling two models: a double pendulum (single stance phase) and a spring-mass model (double stance). Realistic simulations for different Parkinson disease stages are then obtained by integrating to the model a set of trunk bending patterns, learned from real patients. The proposed model was compared with the CoG of real Parkinson gaits in stages 2, 3, 4 achieving a correlation coefficient of 0.88, 0.92 and 0.86, respectively.

Cardiac magnetic resonance imaging (cMRI) is an useful tool in diagnosis, prognosis and research since it functionally tracks the heart structure. Although useful, this imaging technique is limited in spatial resolution because heart is a constant moving organ, also there are other non controled conditions such as patient movements and volumetric changes during apnea periods when data is acquired, those conditions limit the time to capture high quality information. This paper presents a very fast and simple strategy to reconstruct high resolution 3D images from a set of low resolution series of 2D images. The strategy is based on an information reallocation algorithm which uses the DICOM header to relocate voxel intensities in a regular grid. An interpolation method is applied to fill empty places with estimated data, the interpolation resamples the low resolution information to estimate the missing information. As a final step a gaussian filter that denoises the final result. A reconstructed image evaluation is performed using as a reference a super-resolution reconstructed image. The evaluation reveals that the method maintains the general heart structure with a small loss in detailed information (edge sharpening and blurring), some artifacts related with input information quality are detected. The proposed method requires low time and computational resources.

Retinal images often suffer from blurring which hinders disease diagnosis and progression assessment. The restoration of the images is carried out by means of blind deconvolution, but the success of the restoration depends on the correct estimation of the point-spread-function (PSF) that blurred the image. The restoration can be space-invariant or space-variant. Because a retinal image has regions without texture or sharp edges, the blind PSF estimation may fail. In this paper we propose a strategy for the correct assessment of PSF estimation in retinal images for restoration by means of space-invariant or space-invariant blind deconvolution. Our method is based on a decomposition in Zernike coefficients of the estimated PSFs to identify valid PSFs. This significantly improves the quality of the image restoration revealed by the increased visibility of small details like small blood vessels and by the lack of restoration artifacts.

The depth of focus (DOF) defines the axial range of high lateral resolution in the image space for object position. Optical devices with a traditional lens system typically have a limited DOF. However, there are applications such as in ophthalmology, which require a large DOF in comparison to a traditional optical system, this is commonly known as extended DOF (EDOF). In this paper we explore Programmable Diffractive Optical Elements (PDOEs), with EDOF, as an alternative solution to visual impairments, especially presbyopia. These DOEs were written onto a reflective liquid cystal on silicon (LCoS) spatial light modulator (SLM). Several designs of the elements are analyzed: the Forward Logarithmic Axicon (FLAX), the Axilens (AXL), the Light sword Optical Element (LSOE), the Peacock Eye Optical Element (PE) and Double Peacock Eye Optical Element (DPE). These elements focus an incident plane wave into a segment of the optical axis. The performances of the PDOEs are compared with those of multifocal lenses. In all cases, we obtained the point spread function and the image of an extended object. The results are presented and discussed.

Echocardiography is a medical imaging technique based on ultrasound signals that is used to evaluate heart anatomy and physiology. Echocardiographic images are affected by speckle, a type of multiplicative noise that obscures details of the structures, and reduces the overall image quality. This paper shows an approach to enhance echocardiography using two processing techniques: temporal compounding and anisotropic diffusion filtering. We used twenty echocardiographic videos that include one or three cardiac cycles to test the algorithms. Two images from each cycle were aligned in space and averaged to obtain the compound images. These images were then processed using anisotropic diffusion filters to further improve their quality. Resultant images were evaluated using quality metrics and visual assessment by two medical doctors. The average total improvement on signal-to-noise ratio was up to 100.29% for videos with three cycles, and up to 32.57% for videos with one cycle.