Search

The article is available at the following address: https://www.pourlascience.fr/sd/medecine/la-fee-ia-au-chevet-des-malades-23684.php; National audience; Aider au diagnostic, au pronostic et au traitement : les objectifs de l’intelligence artificielle appliquée à l’imagerie médicale ne sont pas minces. Après des premiers succès spectaculaires mais coûteux, la recherche sur les algorithmes vise désormais à établir un « jumeau numérique » du patient.

International audience; We introduce Geomstats, an open-source Python toolbox for computations and statistics on nonlinear manifolds, such as hyperbolic spaces, spaces of symmetric positive definite matrices, Lie groups of transformations, and many more. We provide object-oriented and extensively unit-tested implementations. Among others, manifolds come equipped with families of Riemannian metrics, with associated exponential and logarithmic maps, geodesics and parallel transport. Statistics and learning algorithms provide methods for estimation, clustering and dimension reduction on manifolds. All associated operations are vectorized for batch computation and provide support for different execution backends, namely NumPy, PyTorch and TensorFlow, enabling GPU acceleration. This paper presents the package, compares it with related libraries and provides relevant code examples. We show that Geomstats provides reliable building blocks to foster research in differential geometry and statistics, and to democratize the use of Riemannian geometry in machine learning applications. The source code is freely available under the MIT license at http://geomstats.ai.

Heterogeneity and complexity of biomedical data are a crucial pitfall for modeling neurodegeneration in clinical studies. Modern analysis approaches are required to account for measurements' uncertainty and variability, as well as for the typical large dimensionality of biomedical information. Moreover, model interpretability is an imperative requirement for the clinical translation of automated analysis methods.This project addresses these challenges through three major research axes. The first axis concerns the investigation of novel approaches for the probabilistic modeling of spatio-temporal variations in biomedical data. This research leverages on the extension of non-parametric learning methods, such as Gaussian process (GP) regression, to account for high-dimensional and spatio-temporally correlated signals. Furthermore, it introduces a consistent framework for modeling and inference of the dynamical systems describing the potential bio-mechanical properties governing biological data. The second axis contributes to the definition of learning algorithms for modeling the joint variation between heterogeneous information, such as imaging, clinical, and biological data. Firstly, by defining novel validation strategies for multivariate models applied to high-dimensional brain imaging-genetics and cardiac data. Secondly, by introducing novel formulations of probabilistic latent variable models, where latent representations issued from multiple data views (or channels) are constrained to match a common target latent distribution.The last axis represents a novel direction of investigation aimed at reformulating inference in probabilistic models to cope with the constraint of data privacy and security in multi-centric studies. This research direction addresses an important societal problem, for biomedical data hosted in hospitals usually cannot be shared due to privacy and legal issues. This problem is here tackled by extending variational inference in Bayesian models to the setting of federated learning.

Elles décryptent les mammographies, observent les rétines, scrutent les cerveaux et comparent les symptômes pour prédire d’éventuelles maladies. Les IA sont entrées à l’hôpital et dans les labos, où médecins et chercheurs inventent la santé du XXIe siècle.Des prothèses aux implants en passant par les consultations en ligne et les patients numériques, ces avancées vertigineuses interrogent notre rapport au corps. Elles réveillent le fantasme de l’homme augmenté et le désir d’immortalité des transhumanistes.Grâce à l’éclairage des plus grands experts, cette enquête dresse un état de la recherche et des dernières innovations en matière d’intelligence artificielle. Comment préserver le secret médical dans un monde ultra-connecté ? Les médecins vont-ils disparaître ?Les robots sont-ils une réponse à l’isolement ?Les machines nous privent-elles de notre libre arbitre ?

Structural MRI is widely used for investigating brain atrophy in many neurodegenerative disorders, with several research groups developing and publishing techniques to provide quantitative assessments of this longitudinal change. Often techniques are compared through computation of required sample size estimates for future clinical trials. However interpretation of such comparisons is rendered complex because, despite using the same publicly available cohorts, the various techniques have been assessed with different data exclusions and different statistical analysis models. We created the MIRIAD atrophy challenge in order to test various capabilities of atrophy measurement techniques. The data consisted of 69 subjects (46 Alzheimer's disease, 23 control) who were scanned multiple (up to twelve) times at nine visits over a follow-up period of one to two years, resulting in 708 total image sets. Nine participating groups from 6 countries completed the challenge by providing volumetric measurements of key structures (whole brain, lateral ventricle, left and right hippocampi) for each dataset and atrophy measurements of these structures for each time point pair (both forward and backward) of a given subject. From these results, we formally compared techniques using exactly the same dataset. First, we assessed the repeatability of each technique using rates obtained from short intervals where no measurable atrophy is expected. For those measures that provided direct measures of atrophy between pairs of images, we also assessed symmetry and transitivity. Then, we performed a statistical analysis in a consistent manner using linear mixed effect models. The models, one for repeated measures of volume made at multiple time-points and a second for repeated “direct” measures of change in brain volume, appropriately allowed for the correlation between measures made on the same subject and were shown to fit the data well. From these models, we obtained estimates of the distribution of atrophy rates in the Alzheimer's disease (AD) and control groups and of required sample sizes to detect a 25% treatment effect, in relation to healthy ageing, with 95% significance and 80% power over follow-up periods of 6, 12, and 24 months. Uncertainty in these estimates, and head-to-head comparisons between techniques, were carried out using the bootstrap. The lateral ventricles provided the most stable measurements, followed by the brain. The hippocampi had much more variability across participants, likely because of differences in segmentation protocol and less distinct boundaries. Most methods showed no indication of bias based on the short-term interval results, and direct measures provided good consistency in terms of symmetry and transitivity. The resulting annualized rates of change derived from the model ranged from, for whole brain: − 1.4% to − 2.2% (AD) and − 0.35% to − 0.67% (control), for ventricles: 4.6% to 10.2% (AD) and 1.2% to 3.4% (control), and for hippocampi: − 1.5% to − 7.0% (AD) and − 0.4% to − 1.4% (control). There were large and statistically significant differences in the sample size requirements between many of the techniques. The lowest sample sizes for each of these structures, for a trial with a 12 month follow-up period, were 242 (95% CI: 154 to 422) for whole brain, 168 (95% CI: 112 to 282) for ventricles, 190 (95% CI: 146 to 268) for left hippocampi, and 158 (95% CI: 116 to 228) for right hippocampi. This analysis represents one of the most extensive statistical comparisons of a large number of different atrophy measurement techniques from around the globe. The challenge data will remain online and publicly available so that other groups can assess their methods., Highlights • We compared numerous brain atrophy measurement techniques using multiple metrics. • Each participant produced measures on the exact same dataset, blinded to disease. • A central statistical analysis using linear mixed effect models was performed. • Head to head comparisons for each region were performed using sample size estimates. • Brain and ventricle measures were more consistent across groups than for hippocampi.

International audience; A potential overlap between bipolar disorder (BD) and borderline personality disorder (BPD) has been recently proposed. We aimed to assess similarities and differences of brain structural features in BD and BPD. Twenty-six in-patients with BPD, 14 with BD and 40 age and sex-matched healthy controls (HC) underwent structural magnetic resonance (MR). Voxel-based morphometry analysis with Statistical Parametric Mapping (SPM) was used to localize and quantify gray (GM) and white matter (WM) abnormalities in BD and BPD compared to HC and to identify those specifically affected in each patient group (p

Le domaine de l’invention concerne les méthodes de traitement d’images et de simulation pour déterminer des caractéristiques électriques particulières d’un modèle de myocarde. Plus particulièrement, le domaine de l’invention concerne les méthodes de segmentation d’une paroi du myocarde afin d’isoler et caractériser des phénomènes électriques singuliers à l’intérieur d’un myocarde modélisé.

Glioblastomas are among the most common and aggressive primary brain tumors. It is usually treated with a combination of surgical resection, followed with concurrent chemo- and radiotherapy. However, the infiltrative nature of the tumor makes its control particularly challenging. Biophysical model personalization allows one to automatically define patient specific therapy plans which maximize survival rates. In this thesis, we focused on the elaboration of tools to personalize radiotherapy planning. First, we studied the impact of taking into account the vasogenic edema into the planning. We studied a database of patients treated with anti-angiogenic drug, revealing a posteriori the presence of the edema. Second, we studied the relationship between the uncertainty in the tumor segmentation and dose distribution. For that, we present an approach in order to efficiently sample multiple plausible segmentations from a single expert one. Third, we personalized a tumor growth model to seven patients’ MR images. We used a Bayesian approach in order to estimate the uncertainty in the personalized parameters of the model. Finally, we showed how combining a personalized model of tumor growth with a dose response model could be used to automatically define patient specific dose distribution. The promising results of our approaches offer new perspectives for personalized therapy planning., Les glioblastomes comptent parmi les cas les plus répandus et agressifs de tumeurs cérébrales. Ils sont généralement traités avec une combinaison de résection chirurgicale, suivie de chimiothérapie et radiothérapie. Cependant, le caractère infiltrant de la tumeur rend son traitement particulièrement délicat. La personnalisation de modèles biophysiques permet d’automatiser la mise au point de thérapies spécifiques au patient, en maximisant les chances de survie. Dans cette thèse nous nous sommes attachés à élaborer des outils permettant de personnaliser la radiothérapie des glioblastomes. Nous avons tout d’abord étudié l’impact de la prise en compte de l’œdème vasogénique. Notre étude rétrospective se fonde sur une base de donnée de patients traités avec un médicament anti-angiogénique, révélant a posteriori la présence de l’œdème. Ensuite, nous avons étudié le lien entre l’incertitude due à la segmentation de la tumeur et la distribution de la dose. Pour se faire, nous avons mis au point une méthode permettant d’échantillonner efficacement de multiples segmentations réalistes, à partir d’une unique segmentation clinique. De plus, nous avons personnalisé un modèle de croissance tumorale aux images IRM de sept patients. La méthode Bayésienne adoptée permet notamment d’estimer l’incertitude sur les paramètres personnalisés. Finalement, nous avons montré comment cette personnalisation permet de définir automatiquement la dose à prescrire au patient, en combinant le modèle de croissance tumoral avec un modèle de réponse à la dose délivrée. Les résultats prometteurs présentés ouvrent de nouvelles perspectives pour la personnalisation de la radiothérapie des tumeurs cérébrales.

In french; National audience; Different sets of criteria are currently used for the diagnosis of multiple sclerosis (MS). Some are based on clinical features, while others are related to imaging findings. Among the image processing systems, specific criteria include spatial dissemination of lesions in one image or their temporal dissemination in images acquired at different time points. In addition, the evolution of the lesion load can be used to evaluate treatment efficiency in MS clinical research. Consequently, obtaining a precise segmentation of the MS lesion appears to be crucial. In the literature, a number of semi-automated or completely automated approaches have been proposed enabling a reduction of the inter- and intra-expert variability for manual delineations. A comprehensive state-of-the-art classification of the most representative systems is presented here.

La médecine de demain se dessine aujourd'hui notamment grâce aux progrès issus de l'apport de l'informatique et des sciences numériques au traitement de l'imagerie médicale, elle-même en plein essor, aboutissant à un modèle numérique du patient permettant d'assister le diagnostic, le pronostic et la thérapie.

Tracking gliomas dynamics on MRI has became more and more important for therapeutic management. Powerful computational tools have been recently developed in this context enabling in silico growth on a virtual brain that can be matched with real 3D segmented evolution through registration between atlases and patient brain MRI data. In this paper, we provide an extensive review of existing algorithms for the three computational tasks involved in patient-specific tumor modeling: image segmentation, image registration, and in silico growth modelling (with special emphasis on the proliferation-diffusion model). Accuracy and limits of the reviewed algorithms are systematically discussed. Finally applications of these methods for both clinical practice and fundamental research are also discussed.

In emission tomography imaging, respiratory motion causes artifacts in lungs and cardiac reconstructed images, which lead to misinterpretations, imprecise diagnosis, impairing of fusion with other modalities, etc. Solutions like respiratory gating, correlated dynamic PET techniques, list-mode data based techniques and others have been tested, which lead to improvements over the spatial activity distribution in lungs lesions, but which have the disadvantages of requiring additional instrumentation or the need of discarding part of the projection data used for reconstruction. The objective of this study is to incorporate respiratory motion compensation directly into the image reconstruction process, without any additional acquisition protocol consideration. To this end, we propose an extension to the maximum likelihood expectation maximization (MLEM) algorithm that includes a respiratory motion model, which takes into account the displacements and volume deformations produced by the respiratory motion during the data acquisition process. We present results from synthetic simulations incorporating real respiratory motion as well as from phantom and patient data.

International audience; This paper addresses the generalization of Principal Component Analysis (PCA) to Riemannian manifolds. Current methods like Principal Geodesic Analysis (PGA) and Geodesic PCA (GPCA) minimize the distance to a "Geodesic subspace". This allows to build sequences of nested subspaces which are consistent with a forward component analysis approach. However, these methods cannot easily be adaptedto a backward analysis and lack symmetry in the parametrization of the subspaces. We propose in [10] a new and more general type of family of subspaces in manifolds, barycentric subspaces, which are implicitly defined as the locus of points which are weighted means of k + 1 reference points. Depending on the generalization of the mean that we use, we obtain the Frechet / Karcher / Exponential barycentric subspaces (FBS / KBS / EBS). The completion of the last one is called the affine span. These definitions were shown to define locally submanifolds of dimension k.In this paper, we investigate barycentric subspaces in one of the simplest manifold: the sphere. We show that the affine span is a great subsphere in generic conditions, i.e. also a geodesic subspace. This coincidence of spaces is due to the very high symmetry of the sphere. For second order jets, we show that we obtain subspheres of different radii as in the case of principal nested spheres (PNS) analysis. Among the points of the affine span, determining which ones belong to the Karcher barycentric sub-spaces (KBS) turns out to be a surprisingly difficult problem. Practical experiments show that the KBS covers in general only a small portion of the subsphere containing the reference points. This suggests that the affine span might be a much more interesting definition to work with for subspace definition purposes.We finally discuss the use to these barycentric subspaces to generalize PCA on manifolds, a procedure that we name Barycentric Subpsace Analysis (BSA). Like PGA, barycentric subspaces allow the construction of a forward nested sequence of subspaces which contains the Frechet mean. However, the definition also allows the construction of backward nested sequence which may not contain the mean.

Monsieur l’Administrateur,Monsieur le Président d’Inria,Chers collègues et amis,Mesdames, Messieurs, En 2008, dans sa leçon inaugurale, Gérard Berry nous expliquait « pourquoi et comment le monde devient numérique ». Il y évoquait déjà l’impact croissant de l’informatique en médecine, notamment à travers les développements les plus récents de l’imagerie. Aujourd’hui, c’est pour moi un immense honneur d’être invité dans ce lieu prestigieux, à la chaire annuelle d’Informatique et sciences numé...

Current methods to characterize mesenchymal stem cells (MSCs) are limited to CD marker expression, plastic adherence and their ability to differentiate into adipogenic, osteogenic and chondrogenic precursors. It seems evident that stem cells undergoing differentiation should differ in many aspects, such as morphology and possibly also behaviour; however, such a correlation has not yet been exploited for fate prediction of MSCs. Primary human MSCs from bone marrow were expanded and pelleted to form high-density cultures and were then randomly divided into four groups to differentiate into adipogenic, osteogenic chondrogenic and myogenic progenitor cells. The cells were expanded as heterogeneous and tracked with time-lapse microscopy to record cell shape, using phase-contrast microscopy. The cells were segmented using a custom-made image-processing pipeline. Seven morphological features were extracted for each of the segmented cells. Statistical analysis was performed on the seven-dimensional feature vectors, using a tree-like classification method. Differentiation of cells was monitored with key marker genes and histology. Cells in differentiation media were expressing the key genes for each of the three pathways after 21 days, i.e. adipogenic, osteogenic and chondrogenic, which was also confirmed by histological staining. Time-lapse microscopy data were obtained and contained new evidence that two cell shape features, eccentricity and filopodia (= 'fingers') are highly informative to classify myogenic differentiation from all others. However, no robust classifiers could be identified for the other cell differentiation paths. The results suggest that non-invasive automated time-lapse microscopy could potentially be used to predict the stem cell fate of hMSCs for clinical application, based on morphology for earlier time-points. The classification is challenged by cell density, proliferation and possible unknown donor-specific factors, which affect the performance of morphology-based approaches.

Osteoarticular allograft is one possible treatment in wide surgical resections with large defects. Performing best osteoarticular allograft selection is of great relevance for optimal exploitation of the bone databank, good surgery outcome and patient's recovery. Current approaches are, however, very time consuming hindering these points in practice. We present a validation study of a software able to perform automatic bone measurements used to automatically assess the distal femur sizes across a databank. 170 distal femur surfaces were reconstructed from CT data and measured manually using a size measure protocol taking into account the transepicondyler distance (A), anterior-posterior distance in medial condyle (B) and anterior-posterior distance in lateral condyle (C). Intra- and inter-observer studies were conducted and regarded as ground truth measurements. Manual and automatic measures were compared. For the automatic measurements, the correlation coefficients between observer one and automatic method, were of 0.99 for A measure and 0.96 for B and C measures. The average time needed to perform the measurements was of 16 h for both manual measurements, and of 3 min for the automatic method. Results demonstrate the high reliability and, most importantly, high repeatability of the proposed approach, and considerable speed-up on the planning.

International audience; Tumor growth models based on the FisherKolmogorov reactiondiffusion equation (FK) have shown convincing results in reproducing and predicting the invasion patterns of gliomas brain tumors. Diffusion tensor images (DTIs) were suggested to model the anisotropic diffusion of tumor cells in the brain white matter. However, clinical patient-DTIs are expensive and often acquired with low resolution, which compromises the accuracy of the tumor growth models. In this work, we used the traveling wave approximation model to describe the evolution of the visible boundary of the tumor modeled by the FK equation to investigate the impact of replacing the patient DTI by (i) an isotropic diffusion map or (ii) an anisotropic high-resolution DTI atlas formed by averaging DTIs of multiple patients. We quantify the impact of replacing the patient DTI using three metrics: the shape of the simulated glioma, the estimation of the tumor growth parameters, and the prediction performance on clinical cases.

International audience; This paper evaluates five 3D ultrasound tracking algorithms regarding their ability to quantify abnormal deformation in timing or amplitude. A synthetic database of B-mode image sequences modeling healthy, ischemic and dyssynchrony cases was generated for that purpose. This database is made publicly available to the community. It combines recent advances in electromechanical and ultrasound modeling. For modeling heart mechanics, the Bestel-Clement-Sorine electromechanical model was applied to a realistic geometry. For ultrasound modeling, we applied a fast simulation technique to produce realistic images on a set of scatterers moving according to the electromechanical simulation result. Tracking and strain accuracies were computed and compared for all evaluated algorithms. For tracking, all methods were estimating myocardial displacements with an error below one milimeter on the ischemic sequences. The introduction of a dilated geometry was found to have a significant impact on accuracy. Regarding strain, all methods were able to recover timing differences between segments, as well as low strain values. On all cases, radial strain was found to have a low accuracy in comparison to longitudinal and circumferential components.

Co-auteur : (adni) For The Alzheimer'S Disease Neuroimaging Initiative; International audience; Background. The morphology of the brain observed in patients affected by Alzheimer's disease (AD) is the contribution of different biological processes such as the normal aging and the AD-specific pathological matter loss. Being able to differentiate these complementary biological factors is fundamental in order to isolate and quantify the pathological AD-related structural changes, especially at the earliest phase of the disease, at prodromal and preclinical stages. Methods. We chose the ADNI baseline structural MRIs for 37 healthy subjects positive to the CSF Ab42 (Ab+), 86 patients with mild cognitive impairment (MCI) which subsequently converted to AD, 110 stable MCI, and 134 AD patients. For each subject, a "virtual aging" component was defined as the closest point with respect to the longitudinal deformation modeled for the healthy aging of a group of 63 normal subjects negative to the CSF Ab42 [1]. Once removed the aging component, the remaining specific morphological changes were analyzed group-wise, in order to characterize the atrophy patterns at the different clinical stages, and to test their predictive power in encoding the pathological disease progression. Results. Even though the considered groups did not significantly differ for age, the estimated virtual ages increased as the clinical condition get worse, and were significantly higher for the MCI stable, converters, and AD groups when compared to the healthy Aβ- (p

In medical image analysis, geometrical deformations are used to model intersubject variability. In orthopaedic applications, the geometrical variability is usually observable across anatomical scales. For instance, anatomical differences between mandible bones of different patients can be found on a coarse scale, between the entire left or right side, or on a fine scale, between teeth. Each level of granularity has specific regions of interest in clinical applications. The challenge is to connect the geometrical deformations to clinical regions across scales. In this thesis, we present this connection by introducing structured diffeomorphic registration. At the core of our method is the parametrization of geometrical deformations with trees of locally affine transformations describing intersubject variability across scales. In a second step, we statistically model the deformation parameters in a population by formulating a generative statistical model. This model allows us to incorporate deformation statistics as a prior in a Bayesian setting and it enables us to extend the classical sequential coarse to fine registration to a simultaneous optimization of all scales. This kind of group level prior is natural in a polyaffine context, if we assume one configuration of regions that describes an entire group of images with varying transformations for each region. We validate our approach on a wide range of orthopaedic applications: population-based implant design, biomechanical simulations and allograft selection for femur and mandibles. The improved intelligibility for clinicians and accuracy makes our method a good candidate for clinical use., Dans l'analyse d'images médicales, les déformations géométriques sont utilisées pour modéliser la variabilité entre les patients. Dans les applications orthopédiques, la variabilité géométrique est habituellement observable à différentes échelles. Dans le cas des os mandibulaires, par exemple, on observe des différences anatomiques entre le côté gauche et droit sur une échelle grossière, ou entre les dents sur une échelle plus fine. Chaque niveau de granularité contient des régions d'intérêt pour les applications cliniques. La difficulté est de relier les déformations géométriques avec les régions d'intérêt pour chaque type d'échelles. Dans cette thèse, nous présentons cette liaison par l'introduction du recalage difféomorphe et structuré. Le coeur de notre méthode est le paramétrage des déformations géométriques avec des arbres de transformations localement affines qui décrivent la variabilité entre les patients. En second lieu, nous modélisons statistiquement les paramètres de déformations dans une population par la formulation d'un modèle statistique génératif. Cette méthode nous permet d'intégrer des statistiques de déformations comme une probabilité a priori dans un cadre Bayésien et elle nous permet d'étendre le recalage classique d'un schéma grossier à un schéma fin avec une optimisation simultanée pour toutes les échelles. Nous validons notre approche sur plusieurs applications orthopédiques: la conception des implants pour une population, des simulations biomécaniques et la sélection d'allogreffes. L'amélioration de l'intelligibilité pour les cliniciens et de la précision obtenue fait de notre méthode un candidat prometteur pour des usages cliniques.

Co-auteur : (adni) For The Alzheimer'S Disease Neuroimaging Initiative; International audience; Background. The brain atrophy observed in patients affected by Alzheimer's disease (AD) is the contribution of a normal aging process plus an AD-specific pathological matter loss. Being able to differentiate these complementary atrophy factors is fundamental to isolate and quantify the pathological AD-related structural changes, and might represent a reliable disgnostic measure, especially at prodromal stages of the disease. We propose a method based on non rigid-registration to identify the brain atrophy component specific to the clinical condition, after extraction and removal of the morphological changes described by the normal aging process. Methods. We chose the structural MRIs for 37 healthy subjects positive to CSF Ab42 (Ab+), 91 MCI converters to AD, 106 MCI non-converters and 134 Alzheimer's patients. For each subject, the normal aging component ("virtual age") has been estimated with respect to the longitudinal deformation modeled for the healthy aging of a group of 63 normal subjects negative to the CSF Ab42. Once removed the aging component, the remaining morphological changes have been analysed group-wise, in order to characterize the atrophy patterns at the different clinical stages. Results. The virtual age estimated with respect to the healthy Ab- longitudinal deformation is shown in Figure 1. Even though the considered groups did not significantly differ for age, the estimated virtual ages significantly increase as the clinical condition get worse. Once removed the healthy aging component, the average morphological changes specific for the clinical groups are shown in Figure 2. The morphological changes specific for the healthy Ab- are mild and mainly located in the frontal lobe, ventricles and entorhinal cortex. The changes specific for the MCI converters are stronger and map to the frontal lobe, ventricles, temporal poles, entorhinal cortex and hippocampus. The same pattern, although more pronounced, is appreciable for the AD patients. Conclusion. We propose a method to decompose the brain atrophy into complementary components: aging and AD specific. These components identify the different clinical stages, and suggest that more advanced AD stages are associated to both "virtually older" brains and increased specific morphological changes that are not related to the normal aging. These results provide new insights that can lead to new understandings of the AD dynamics, and to novel diagnostic techniques for the early detection of the disease. no abstract

1st International Symposium on Deep Brain Connectomics, Clermont-Ferrand, France; no abstract

medInria is a free medical imaging software developed at Inria, which aims at providing clinicians with state-of-the-art algorithms dedicated to medical image processing and visualization. Efforts have been made to simplify the user interface, while keeping high-level algorithms. In this particular article, we will concentrate on its use in preoperative preparation for cardiac interventions, and how we handle the difficulties arising from the lack of standard format for data types such as meshes or fibers, the absence of a common programming interface for data processing algorithms, notably registration, and the issues of visualisation where display conventions would be beneficial.

International audience; A fully automatic algorithm is presented for the automatic segmentation of gliomas in 3D MR images. It builds on the discriminative random decision forest framework to provide a voxel-wise probabilistic classi cation of the volume. Our method uses multi-channel MR intensi- ties (T1, T1C, T2, Flair), spatial prior and long-range comparisons with 3D regions to discriminate lesions. A symmetry feature is introduced ac- counting for the fact that gliomas tend to develop in an asymmetric way. Quantitative evaluation of the data is carried out on publicly available labeled cases from the BRATS Segmentation Challenge 2012 dataset and demonstrates improved results over the state of the art.

International audience; In computer-assisted orthopaedic surgery, recovering three-dimensional patient-specific anatomy from incomplete information has been focus of interest due to several factors such as less invasive surgical procedures, reduced radiation doses, and rapid intra-operative updates of the anatomy. The aim of this paper is to report results obtained combining statistical shape modeling and multivariate regression techniques for predicting bone shape from clinically and surgically relevant predictors, including sparse observations of the bone surface but also morphometric and anthropometric information. Different state of the art methods such as partial least square regression, principal component regression, canonical correlation analysis, and non-parametric kernel-based regression are compared. Clinically relevant surrogate variables and combinations are investigated on a database of 142 femur and 154 tibia shapes obtained from CT images. The results are evaluated using cross validation to quantify the prediction error. The proposed approach enables to characterize the added value of different predictors in a quantitative and localized fashion. Results indicate that complementary sources of information can be efficiently exploited to improve the accuracy of shape prediction.

We present an original method for the online blind calibration of non-uniform photodetectors. The disparity of the detectors may arise from both irregular spatial arrangement and distinct slowly timevarying photometric transfer functions. As natural images are mostly continuous, the signal collected by neighboring detectors is strongly correlated over time. The core idea of our method is to translate the calibration problem into relative pairwise calibrations between neighboring detectors followed by the regularized inversion of a system akin to gradientbased surface recovery. From our blind calibration procedure, we design an online blind calibration pipeline compatible with clinical practice. On- line blind calibration is proved to be statistically better than standard offline calibration for reconstructing endomicroscopy sequences.

International audience; The morphology observed in the brain of patients a ected by Alzheimer's disease (AD) is the contribution of di erent biological processes such as the normal aging and the AD-speci c pathological matter loss. The ability to di erentiate these complementary biological factors is fundamental in order to reliably evaluate the pathological AD related structural changes, especially at the earliest phase of the disease, at prodromal and pre-clinical stages. We propose a method based on non-rigid registration to estimate the di erent contributions of these complementary factors, and to identify the brain structural changes which are speci c for the pathological component. The experimental results provide a description of the anatomical changes observed across the AD time span: normal aging, normal aging at risk, conversion to MCI and latest AD stages. More advanced AD stages are associated to "virtually older" brains, and to increased speci c morphological changes that are not related to the normal aging. These results provide new insights that can lead to new understandings of the AD dynamics, and to novel techniques for the modeling and the early detection of the disease.

The aim of the present thesis is to provide an e ffective computational framework for the analysis and quantifi cation of the longitudinal structural changes in Alzheimer's disease (AD). The framework is based on the diffeomorphic non-rigid registration parameterized by stationary velocity fields (SVFs), and is hierachically developed to account for the diff erent levels of variability which characterize the longitudinal observations of T1 brain magnetic resonance images (MRIs). We developed an effi cient and robust method for the quantifi cation of the structural changes observed between pairs of MRIs. For this purpose, we propose the LCC-Demons registration framework which implements the local correlation coeffi cient as similarity metric, and we derived consistent and numerically stable measures of volume change and boundary shift for the regional assessment of the brain atrophy. In order to consistently analyze group-wise longitudinal evolutions, we then investigated the parallel transport of subject-specifi c deformation trajectories across di fferent anatomical references. Based on the SVF parametrization of diffeomorphisms, we relied on the Lie group theory to propose new and effective strategies for the parallel transport of SVFs, with particular interest into the practical application to the registration setting. These contributions are the basis for the defi nition of qualitative and quantitative analysis for the pathological evolution of AD. We proposed several analysis frameworks which addressed the di fferentiation of pathological evolutions between clinical populations, the statistically powered evaluation of regional volume changes, and the clinical diagnosis at the early/prodromal disease stages.

The goal of this thesis is to provide automatic segmentation tools based on the use of anatomical atlases. The proposed algorithms are applied and evaluated in the context of radiotherapy of head and neck cancers : segmentation of the organs at risk and the lymph node levels of the neck for treatment planning, and segmentation of the teeth to assess the risk level associated to a given post-irradiation dental care. We first address the construction of an average atlas from a database of manually delineated images considered as reference atlases. In particular, we propose several algorithms to estimate an average segmentation from incomplete segmentations that comprise of missing contours. In order to get over the limits inherent in the average atlas, and in particular its di culty to cope with the inter-individual anatomical variability, we propose to build several average atlases after strati fication of the database into several sub-groups, and to use for each patient the most appropriate average atlas. Then, we present several approaches to select and fuse the most relevant atlases of the database for each patient. We propose global, regional, and local approaches for atlas selection, and an unbiaised protocol to assess them. We also propose several methods for the atlas fusion step, and in particular the construction of a patient-specifi c average atlas from the most similar atlases selected on each region. We show that the selection and fusion of atlases enable a signi ficant improvement compared to the use of the average atlas. Most of this work was integrated into the software ISOgray commercialised by DOSIsoft., L'objectif de cette thèse est de fournir des outils de segmentation automatique basés sur l'utilisation d'atlas anatomiques. Les algorithmes proposés sont appliqués et évalués dans le cadre de la radiothérapie de la région ORL : segmentation des organes à risque et des aires ganglionnaires pour la plani cation du traitement, et segmentation des dents pour l'évaluation des risques liés aux soins dentaires post-irradiation. Nous abordons d'abord la construction d'un atlas moyen à partir d'une base d'images manuellement segmentées considérées comme atlas de référence. Nous proposons en particulier plusieurs algorithmes permettant d'estimer une segmentation moyenne à partir de segmentations incomplètes comportant des contours manquants. A n de dépasser les limites inhérentes à l'atlas moyen unique, et notamment ses diffi cultés à gérer la variabilité anatomique inter-individuelle, nous proposons de construire plusieurs atlas moyens à partir d'une stratifi cation de la base d'atlas, et d'utiliser pour chaque patient l'atlas moyen le plus adapté parmi ceux disponibles. Nous présentons ensuite plusieurs approches permettant de sélectionner puis de fusionner pour chaque patient les atlas de la base les plus pertinents. Nous proposons des approches globales, régionales et locales de sélection d'atlas et un protocole non biaisé pour les évaluer. Nous proposons aussi diff érents schémas pour la fusion d'atlas, notamment la construction d'un atlas moyen composite adapté au patient. Nous montrons que la sélection et la fusion d'atlas apportent une amélioration signi ficative par rapport à l'atlas moyen. La plupart de ces travaux ont été intégrés dans le logiciel ISOgray de DOSIsoft

Memantine is an approved symptomatic treatment for moderate to severe Alzheimer's disease that reduces the excitotoxic effects of hyperactive glutamatergic transmission. However, the exact mechanism of the effect of memantine in Alzheimer's disease patients is poorly understood. Importantly, the default mode network (DMN), which plays a key role in attention, is hypoactive in Alzheimer's disease and is under glutamatergic control.To assess the effect of memantine on the activity of the DMN in moderate to severe Alzheimer's disease.Functional magnetic resonance imaging (MRI) data from 15 patients with moderate to severe Alzheimer's disease, seven treated with memantine (mean ± SD age 77 ± 8 years, mean ± SD Mini-Mental State Examination [MMSE] score 16 ± 5) and eight with placebo (mean ± SD age 76 ± 6 years, mean ± SD MMSE score 13 ± 1), were acquired at baseline (T0) and after 6 months of treatment (T6). Resting state components were extracted after spatial normalization in individual patients with independent component analysis. The consistency of the components was assessed using ICASSO and the DMN was recognized through spatial correlation with a pre-defined template. Voxel-based statistical analyses were performed to study the change in DMN activity from T0 to T6 in the two groups.At T0, the two groups showed similar DMN activity except in the precuneus and cuneus, where the patients who started treatment with memantine had slightly greater activity (p0.05 corrected for familywise error [FWE]). The prospective comparison between T0 and T6 in the treated patients showed increased DMN activation mapping in the precuneus (p0.05, FWE corrected), while the prospective comparison in the untreated patients did not show significant changes. The treatment × time interaction term was significant at p0.05, FWE corrected.The results suggest a positive effect of memantine treatment in patients with moderate to severe Alzheimer's disease, resulting in an increased resting DMN activity in the precuneus region over 6 months. Future studies confirming the present findings are required to further demonstrate the beneficial effects of memantine on the DMN in Alzheimer's disease.

International audience; Statisticalmodelshavebeenrecentlyintroducedincomputational orthopaedics to investigate the bone mechanical properties across several populations. A fundamental aspect for the construction of statistical models concerns the establishment of accurate anatomical correspondences among the objects of the training dataset. Various methods have been proposed to solve this problem such as mesh morphing or image registration algorithms. The objective of this study is to compare a mesh-based and an image-based statistical appearance model approaches for the creation of finite element(FE) meshes. A computer tomography (CT) dataset of 157 human left femurs was used for the comparison. For each approach, 30 finite element meshes were generated with the models. The quality of the obtained FE meshes was evaluated in terms of volume, size and shape of the elements. Results showed that the quality of the meshes obtained with the image-based approach was higher than the quality of the mesh-based approach. Future studies are required to evaluate the impact of this finding on the final mechanical simulations.

International audience; Biophysical models are increasingly used for medical applications at the organ scale. However, model predictions are rarely associated with a confidence measure although there are important sources of uncertainty in computational physiology methods. For instance, the sparsity and noise of the clinical data used to adjust the model parameters (personalization), and the difficulty in modeling accurately soft tissue physiology. The recent theoretical progresses in stochastic models make their use computationally tractable, but there is still a challenge in estimating patient-specific parameters with such models. In this work we propose an efficient Bayesian inference method for model personalization using polynomial chaos and compressed sensing. This method makes Bayesian inference feasible in real 3D modeling problems. We demonstrate our method on cardiac electrophysiology. We first present validation results on synthetic data, then we apply the proposed method to clinical data. We demonstrate how this can help in quantifying the impact of the data characteristics on the personalization (and thus prediction) results. Described method can be beneficial for the clinical use of personalized models as it explicitly takes into account the uncertainties on the data and the model parameters while still enabling simulations that can be used to optimize treatment. Such uncertainty handling can be pivotal for the proper use of modeling as a clinical tool, because there is a crucial requirement to know the confidence one can have in personalized models.

Follow-up imaging studies require the population-wise evaluation of the anatomical changes over time for specific clinical groups. A typical example is the analysis of structural brain changes with aging in Alzheimer's disease. The longitudinal morphological changes for a specific subject can be evaluated through the non-rigid registration. However, to perform a longitudinal group-wise analysis, the subjectspecific longitudinal trajectories need to be transported in a common reference. The analysis of existing transport methods suggests that the inconsistency of the discrete approximations with composition could drown the small longitudinal deformations into the large inter-subject variability. Instead of approximating a continuous parallel transport in an inconsistent discrete setting, we propose to rely on a carefully designed discrete construction that intrinsically respects all the symmetries on the problem: the Schild's Ladder procedure. We show that the Schild's ladder procedure can be implemented extremely efficiently with great stability in the framework of diffeomorphisms parameterized by stationary velocity fields (SVF). The proposed algorithm proves to overcome other methods in terms of consistency with respect to the resampling of most scalar measures of deformations, which opens the way to stable and hopefully more powerful group-wise statistics of longitudinal deformations, Les 'etudes longitudinales en imagerie requi'erent l''evaluation des changements anatomiques au cours du temps sur des populations repr'esentatives de groupes cliniques. Un exemple typique est l''evaluation des modifications structurales du cerveau avec le vieillissement dans la maladie d'Alzheimer. Pour un sujet donn'e, les changements morphologiques longitudinaux peuvent ˆetre 'evalu'es grˆace au recalage non-rigide. Cependant, ces trajectoires longitudinales doivent ˆetre replac'ees dans un r'ef'erentiel commun pour ˆetre analys'ees au niveau du groupe. Or si le transport de fonctions scalaires est clair, le transport de vecteurs ou de d'eformations est moins bien d'efini. L'analyse des m'ethodes existantes sugg'ere que l'inconsistance des approximations discr'etes avec la composition de transformations puisse noyer les d'eformation longitudinale relativement faible dans la tr'es grande variabilit'e inter-sujets. Au lieu d'approximer le transport parall'ele continu avec des sch'emas discrets inconsistants, nous proposons d'utiliser une construction discr'ete qui respecte intrins'equement toutes les sym'etries du probl'eme : l''echelle de Schild. Nous proposons une reformulation particuli'erement simple dans le cas des cadre de diff'eomorphismes param'etr'es par des champs de vecteurs stationnaires. Les deux algorithmes impl'ementant cette construction s'av'erent tr'es efficace et plus consistante que les autres m'ethodes avec le r'e'echantillonage de la plupart des mesures scalaires r'esumant la d'eformation longitudinale. Ceci ouvre la voie des statistiques plus stables sur la transformation de groupe et des statistiques de groupe longitudinales plus puissantes.

International audience; In this paper we present the first Content-Based Image Retrieval (CBIR) framework in the field of in vivo endomicroscopy, with applications ranging from training support to diagnosis support. We propose to adjust the standard Bag-of-Visual-Words method for the retrieval of endomicroscopic videos. Retrieval performance is evaluated both indirectly from a classification point-of-view, and directly with respect to a perceived similarity ground truth. The proposed method significantly outperforms, on two different endomicroscopy databases, several state-of-the-art methods in CBIR. With the aim of building a self-training simulator, we use retrieval results to estimate the interpretation difficulty experienced by the endoscopists. Finally, by incorporating clinical knowledge about perceived similarity and endomicroscopy semantics, we are able: 1) to learn an adequate visual similarity distance and 2) to build visual-word-based semantic signatures that extract, from low-level visual features, a higher-level clinical knowledge expressed in the endoscopist own language.

International audience; Purpose To propose an automatic atlas-based segmentation framework of the dental structures, called Dentalmaps, and to assess its accuracy and relevance to guide dental care in the context of intensity-modulated radiotherapy. Methods and Materials A multi-atlas-based segmentation, less sensitive to artifacts than previously published head-and-neck segmentation methods, was used. The manual segmentations of a 21-patient database were first deformed onto the query using nonlinear registrations with the training images and then fused to estimate the consensus segmentation of the query. Results The framework was evaluated with a leave-one-out protocol. The maximum doses estimated using manual contours were considered as ground truth and compared with the maximum doses estimated using automatic contours. The dose estimation error was within 2-Gy accuracy in 75% of cases (with a median of 0.9 Gy), whereas it was within 2-Gy accuracy in 30% of cases only with the visual estimation method without any contour, which is the routine practice procedure. Conclusions Dose estimates using this framework were more accurate than visual estimates without dental contour. Dentalmaps represents a useful documentation and communication tool between radiation oncologists and dentists in routine practice. Prospective multicenter assessment is underway on patients extrinsic to the database.

International audience; PURPOSE To evaluate the correlation between scar as depicted by delayed enhancement (DE) MRI and the electrophysiological arrhythmogenic substrate in scar-related ventricular tachycardia (VT). METHOD AND MATERIALS 10 patients (4 ICM, 3 myocarditis, 2 ARVC, 1 non-ischemic DCM) underwent DE MRI at high spatial resolution (voxel size 0.625x0.625x2.5mm) prior to VT ablation. The endocardium and epicardium were manually segmented. Scar was automatically segmented on myocardial histogram with a threshold set at 3SD above mean signal of remote myocardium. From the segmented images, a digital 3D model was computed and imported in 3D mapping systems. During the mapping/ablation procedure, this model was co-registered in real time to the electrophysiological mapping geometry. Endocardial and epicardial contact mapping was performed at high density (>500 sites per map) during sinus rhythm (SR) and VT. The electrophysiological VT substrate (abnormal/late activity during SR, critical isthmus during VT) was correlated to DE MRI data. RESULTS The integration of MRI data enabled a 3D visualization of the endocardium, epicardium and scar during the mapping/ablation procedure. VT was mapped in 7/10 patients. Critical isthmuses located within scar areas in 4 patients (2 myocarditis, 1 ICM, 1 ARVC), and at the border zone (

During the last past years, significant progress in Medical Image Analysis, in biomathematics and biophysics have led to development of the first personalized digital cardiac models. These digital models are personalized which means they can reproduce the anatomy and physiology of specific patients. They allow the quantitative analysis of the organ function and the simulation of some therapies to evaluate their expected benefit. In this article we describe some recent research work done on these topics in our project team Asclepios at Inria, in collaboration with other Inria teams (Macs, Reo, Sisyphe) and external academic, clinical and industrial partners. If a number of challenges in mathematics and informatics still have to be solved before such personalized digital cardiac models can be used in current clinical practice, these first results announce a new generation of tools in digital medicine which can contribute more widely to a more preventive and more predictive personalized medicine., Au cours de ces dernières années, des progrès importants dans l'analyse informatique des images médicales et dans la modélisation biomathématique et biophysique ont rendu possible la construction des premiers modèles numériques et personnalisés du cœur humain. Ces modèles informatiques sont personnalisés car ils reproduisent l'anatomie et la physiologie de patients spécifiques. Ils permettent d'analyser et de quantifier le fonctionnement de l'organe et de simuler certainesthérapies pour en évaluer le bénéfice espéré. Dans cet article nous décrivons des travaux de recherche récents réalisés sur ce thème au sein de l'équipe projet Asclepios à l'Inria, en collaboration avec d'autres équipes Inria (Macs, Reo, Sisyphe) et des partenaires extérieurs académiques, cliniques et industriels. Si de grands défis en modélisation informatique et mathématique doivent encore être relevés avant une utilisation clinique courante du cœur numérique personnalisé, ces premiers résultats annoncent une nouvelle génération d'outils de médecine numérique pouvant contribuer plus largement à une médecine personnalisée plus préventive et plus prédictive.

VPH European Conference; International audience; Background. The brain of patients with Alzheimer's disease undergoes changes starting from many years before the development of the first clinical symptoms. Large prospective neuroimaging datasets provide data to study the structural changes over time in health and disease. The evaluation of the longitudinal changes in the brain poses several challenges concerning the consistency of the measurements at both spatial and temporal level and the development of a reliable general model to adapt to the subject-specific temporal evolution. Objectives. 1) To use deformation based morphometry to develop a unique framework to consistently quantify the changes in the brain, at both local (voxel) and regional level. 2) To describe the longitudinal changes in the brain affected by Alzheimer's under a multilevel approach, where the subject-specific structural changes at each time point are integrated to describe a subject-specific trajectory for the changes in time composed by coherent serial deformations. Methods. We employed the longitudinal scans for a group of 8 Alzheimer's patients from the ADNI dataset. For each subject we measured the deformations to match the follow-up scans to the baseline using the Symmetric Log-Domain Demons Algorithm. The degree of expansion/contraction in the whole brain was assessed at two different levels within the same framework * Global change at regional level (Shift of the boundaries): computing the flux of the deformation across the brain's surface. * Local change at voxel level (Infinitesimal volume change): point evaluation of the deformation's Jacobian determinant. For each subject, the sequence of deformations at each time point was used to define a subject-specific prior longitudinal deformation and the prior was used to re-estimate the deformations as a weighted average between spatial and temporal information. Results. The correlation between manual derived brain volume changes and the measure of whole brain changes derived from the deformation field was of 0.899. The temporal regularization procedure generally led to more linear trends for the subject-specific changes in time and increased the correlation with manual derived measures reaching the maxima of 0.922 when the weight of the temporal information was half of the spatial. Conclusions. Deformation based morphometry provides a framework to evaluate the changes in the brain along the time consistently along the spatial dimension, from the voxel to the regional level, and consistently along the temporal dimension, regularising the longitudinal estimation of the deformations.

International audience; Diffusion tensor images (DTIs) provide information about deep white matter anatomy that structural magnetic resonance images typically fail to resolve. Non-linear registration of DTIs provides a way to capture the deformations of these structures that would otherwise go unobserved. Here we use an existing method that fully incorporates a useful vector space parameterization of diffeomorphisms, thereby allowing simple and well defined calculation of deformation statistics. An initial analysis of the statistics produced by registration of a group of 37 HIV/AIDS patients illustrates principal modes of deformation that are anatomically meaningful and that corroborate with previous findings. The registration method is developed by incorporating these modes into a statistical regularization criterion. Even though initial results suggest this new criterion over-constrains the registration method, we discuss plausible ways to address this.

International audience

1936-959X (Electronic) 0195-6108 (Linking) Controlled Clinical Trial Journal Article Research Support, Non-U.S. Gov't; International audience; BACKGROUND AND PURPOSE: The physiopathologic bases underlying the signal intensity changes and reduced diffusibility observed in prion diseases (TSEs) are still poorly understood. We evaluated the interest of MRS combined with DWI both as a diagnostic tool and a way to understand the mechanism underlying signal intensity and ADC changes in this setting. MATERIALS AND METHODS: We designed a prospective study of multimodal MR imaging in patients with suspected TSEs. Forty-five patients with a suspicion of TSE and 11 age-matched healthy volunteers were included. The MR imaging protocol included T1, FLAIR, and DWI sequences. MRS was performed on the cerebellum, pulvinar, right lenticular nucleus, and frontal cortex. MR images were assessed visually, and ADC values were calculated. RESULTS: Among the 45 suspected cases, 31 fulfilled the criteria for probable or definite TSEs (19 sCJDs, 3 iCJDs, 2 vCJDs, and 7 genetic TSEs); and 14 were classified as AltDs. High signals in the cortex and/or basal ganglia were observed in 26/31 patients with TSEs on FLAIR and 29/31 patients on DWI. In the basal ganglia, high DWI signals corresponded to a decreased ADC. Metabolic alterations, increased mIns, and decreased NAA were observed in all patients with TSEs. ADC values and metabolic changes were not correlated; this finding suggests that neuronal stress (vacuolization), neuronal loss, and astrogliosis do not alone explain the decrease of ADC. CONCLUSIONS: MRS combined with other MR imaging is of interest in the diagnosis of TSE and provides useful information for understanding physiopathologic processes underlying prion diseases.

International audience; The registration of thoracic images is a challenging problem with essential clinical applications such as radiotherapy and diagnosis. In the context of the EMPIRE10 challenge, we brie y introduce a general robust and e cient algorithm to register automatically any type of scalar images (CT, MRI, ...) on virtually any location (brain, thorax, ...). Although fully automatic and generic, the proposed algorithm reached the 17th place in the EMPIRE challenge over 34 algorithms evaluated. Moreover, we have since optimized further the parameter set used for the challenge and we demonstrate the ability of the algorithm to recover much better large displacements of the lungs boundaries.

International audience; The registration of thoracic images is a common but still challenging problem with critical clinical applications (e.g. radiotherapy and diagnosis). In the context of the EMPIRE10 challenge, we briefly introduce in this paper our registration method based on the di ffeomorphic demons algorithm. Although fully automatic and generic (applies to a large variety of images such as brain or thoracic CT scans), the proposed method appears to be a very e fficient registration method.

None, Aucun

International audience; no abstract

This thesis enters on the development of a point-based statistical shape model relying on correspondence probabilities in a sound mathematical framework. Further focus lies on the integration of the model into a segmentation method where a novel approach is taken by combining an explicitly represented shape prior with an implicitly represented segmentation contour. In medical image analysis, the notion of shape is recognized as an important feature to distinguish and analyse anatomical structures. The modeling of shape realized by the concept of statistical shape models constitutes a powerful tool to facilitate the solutions to analysis, segmentation and reconstruction problems. A statistical shape model tries to optimally represent a set of segmented shape observations of any given organ via a mean shape and a variability model. A fundamental challenge in doing statistics on shapes lies in the determination of correspondences between the shape observations. The prevailing assumption of one-to-one point correspondences seems arguable due to uncertainties of the shape surface representations as well as the general di fficulty of pinpointing exact correspondences. In this thesis, the following solution to the point correspondence problem is derived: For all point pairs, a correspondence probability is computed which amounts to representing the shape surfaces by Mixtures of Gaussians. This approach allows to formulate the model computation in a generative framework where the shape observations are interpreted as randomly generated by the model. Based on that, the computation of the model is then treated as an optimization problem. An algorithm is proposed to optimize for model parameters and observation parameters through a single maximum a posteriori criterion which leads to a mathematically sound and unified framework. The method is evaluated and validated in a series of experiments on synthetic and real data. To do so, adequate performance measures and metrics are defined based on which the quality of the new model is compared to the qualities of a classical point-based model and of an established surface-based model that both rely on one-to-one correspondences. A segmentation algorithm is developed which employs the a priori shape knowledge inherent in the statistical shape model to constrain the segmentation contour to probable shapes. An implicit segmentation sheme is chosen instead of an explicit one, which is beneficial regarding topological exibility and implementational issues. The mathematically sound probabilistic shape model enables the challenging integration of an explicit shape prior into an implicit segmentation scheme in an elegant formulation. A maximum a posteriori estimation is developed of a level set function whose zero level set best separates the organ from the background under a shape constraint introduced by the model. This leads to an energy functional which is minimized with respect to the level set using an Euler-Lagrangian equation. Since both the model and the implicitly defined contour are well suited to represent multi-object shapes, an extension of the algorithm to multi-object segmentation is developed which is integrated into the same probabilistic framework. The novel method is evaluated on kidney and hip joint segmentation.

Despite advances in both medical image analysis and intracardiac electrophysiological mapping technology, the understanding of cardiac mechano-electrical coupling is still incomplete. This knowledge is of high interest since it would help estimating the cardiac electrophysiology function from the analysis of widely available cardiac images, such as 3D echocardiography. This is important, for example, in the evaluation of the cardiac resynchronization therapy (CRT) where the placement and tuning of the pacemaker leads plays a crucial role in the outcome of the therapy. This paper proposes a method to estimate activation times of myocardium using a cardiac electromechanical model. We use Kernel Ridge Regression to find the relationship between the kinematic descriptors (strain and displacement) and the contraction force caused by the action potential propagation. This regression model is then applied to two 3D echocardiographic sequences from a patient, one in sinus rhythm and the other one with left ventricle pacing, for which strains and displacements have been estimated using incompressible diffeomorphic demons for non-rigid registration.

This paper presents a method for nonlinear registration of images, where there exists no one-to-one correspondence in parts of the image. Such a situation occurs for instance in the case where an atlas of normal anatomy shall be matched to pathological data, such as tumors, resections or lesions. Our idea is to use local confidence weights and to model pathological regions with zero confidence. We integrate this concept into the efficient and publicly available diffeomorphic demons registration framework. Finally, we show that this process better captures deformations in high-confidence regions than without using the proposed modification. Furthermore, it is easy to implement and runs faster than previous approaches.

no abstract

Clinical trials of disease modifying drugs for Alzheimer's disease (AD) in patients with mild cognitive impairment (MCI) might benefit from enrichment with true AD cases. Four hundred five MCI patients (143 converters and 262 nonconverters to AD within 2 years) of the Alzheimer's disease Neuroimaging Initiative (ADNI) were used. Markers for enrichment were hippocampal atrophy on magnetic resonance (MRI), temporoparietal hypometabolism on FDG PET, cerebrospinal fluid (CSF) biomarkers (Abeta42, tau, and phospho-tau), and cortical amyloid deposition (11C-PIB positron emission tomography (PET)). Two separate enrichment strategies were tested to A) maximize the proportion of MCI converters screened in, and B) minimize the proportion of MCI converters screened out. Based on strategy A, when compared with no enrichment and ADAS-Cog as an outcome measure (sample size of 834), enrichment with 18F-FDG PET and hippocampal volume lowered samples size to 260 and 277 cases per arm, but at the cost of screening out 1,597 and 434 cases per arm. When compared with no enrichment and clinical dementia rating (CDR-SOB) as an outcome measure (sample size of 674), enrichment with hippocampal volume and Abeta42 lowered sample sizes to 191 and 291 cases per arm, with 639 and 157 screened out cases. Strategy B reduced the number of screened out cases (740 for [11C]-PIB PET, 101 hippocampal volume, 82 ADAS-COG and 330 for [18F]-FDG PET) but at the expense of decreased power and a relative increase size (740 for [11C]-PIB PET, 676 for hippocampal volume, 744 for ADAS-Cog, and 517 for [18F]-FDG PET). Enrichment comes at the price of an often relevant proportion of screened out cases, and in clinical trial settings, the balance between enrichment of screened in and loss of screened out patients should be critically discussed.