Search

International audience; ObjectivesThe accurate preoperative segmentation of the uterus and uterine fibroids from magnetic resonance images (MRI) is an essential step for diagnosis and real-time ultrasound guidance during high-intensity focused ultrasound (HIFU) surgery. Conventional supervised methods are effective techniques for image segmentation. Recently, semi-supervised segmentation approaches have been reported in the literature. One popular technique for semi-supervised methods is to use pseudo-labels to artificially annotate unlabeled data. However, many existing pseudo-label generations rely on a fixed threshold used to generate a confidence map, regardless of the proportion of unlabeled and labeled data.Materials and MethodsTo address this issue, we propose a novel semi-supervised framework called Confidence-based Threshold Adaptation Network (CTANet) to improve the quality of pseudo-labels. Specifically, we propose an online pseudo-labels method to automatically adjust the threshold, producing high-confident unlabeled annotations and boosting segmentation accuracy. To further improve the network's generalization to fit the diversity of different patients, we design a novel mixup strategy by regularizing the network on each layer in the decoder part and introducing a consistency regularization loss between the outputs of two sub-networks in CTANet.ResultsWe compare our method with several state-of-the-art semi-supervised segmentation methods on the same uterine fibroids dataset containing 297 patients. The performance is evaluated by the Dice similarity coefficient, the precision, and the recall. The results show that our method outperforms other semi-supervised learning methods. Moreover, for the same training set, our method approaches the segmentation performance of a fully supervised U-Net (100% annotated data) but using 4 times less annotated data (25% annotated data, 75% unannotated data).ConclusionExperimental results are provided to illustrate the effectiveness of the proposed semi-supervised approach. The proposed method can contribute to multi-class segmentation of uterine regions from MRI for HIFU treatment.

Clinical interventional hemodynamic studies quantify the ventricular function from two-dimensional (2-D) X-ray projection images without having enough information of the actual three-dimensional (3-D) shape of this cardiac cavity. This paper reports a left ventricle 3-D reconstruction method from two orthogonal angiographic projections. This investigation is motivated by the lack of information about the actual 3-D shape of the cardiac cavity. The proposed algorithm works in 3-D space and considers the oblique projection geometry associated with the biplane image acquisition equipment. The reconstruction process starts by performing an approximate reconstruction based on the Cylindrical Closure Operation and the Dempster-Shafer theory. This approximate reconstruction is appropriately deformed in order to match the given projections. The deformation procedure is carried out by an iterative process that, by means of the Dempster-Shafer and the fuzzy integral theory, combines the information provided by the projection error and the connectivity between voxels. The performance of the proposed reconstruction method is evaluated by considering first the reconstruction of two 3-D binary databases from two orthogonal synthetized projections, obtaining errors as low as 6.48%. The method is then tested on real data, where two orthogonal preprocessed angiographic images are used for reconstruction. The performance of the technique, in this case, is assessed by means of the projection error, whose average for both views is 7.5%. The reconstruction method is also tested by performing the 3-D reconstruction of a ventriculographic sequence throughout an entire cardiac cycle.

This short survey of moment invariants points out interesting clues. Moments offer a sound theoretical framework for solving the generic problems encountered in many imaging applications. The diverse families of orthogonal moments provide the flexibility that may be required to face a particular target. However, they have to satisfy the time computation constraints that are inherent in many applications. Suk and Flusser proposed a solution of simultaneously dealing with affine transformation and blur (with centrosymmetric PSF) for pattern recognition, template matching, and image registration. Flusser and Zivota suggested a set of combined moments that are invariant to both rotation and blurring. Based on complex moments, Liu and Zhang derived a subset of moment features that are not affected by image blurring and geometric transformation such as translation, scale, and rotation. All these works, however, point out the problems related to the number of invariants to be selected, the choice of the size of the region of interest where moments are computed, and the dependence with object features (i.e., symmetry).

In the field of epilepsy, the analysis of stereoelectroencephalographic (SEEG) signals recorded with depth electrodes provides major information on interactions between brain structures during seizures. A methodology of comparing SEEG seizure recordings is applied in 4 patients suffering from temporal lobe epilepsy. It proceeds in 3 steps: (i) segmentation of SEEG signals, (ii) characterization and labeling of segments and (iii) comparison of observations coded as sequences of symbol vectors. The third step is based on vectorial extension of Wagner and Fischer's algorithm to first, quantify similarities between observations and second, extract invariant information, referred to as spatio-temporal signatures. These are automatically extracted by the algorithm without the need to make a priori assumptions on the ‘patterns’ to be searched for. Theoretical results show that two observations of non-equal duration can be matched by deforming the first one (using insertion/deletion operations on vectors) to optimally fit the second, under a minimal cost constraint. Clinical results show that the study brings objective results on reproducible mechanisms occurring during seizures: for a given patient, quantified descriptions of seizure periods are compared and similar ictal patterns, or signitures, are extracted from SEEG signals. Some of these signatures (particularly those containing spikes, spike-and-waves, slow waves and rapid discharges) are relevant: they seem to reflect reproducible propagation schemes whose analysis may help in the understanding of epileptogenic networks.

International audience; The Health Directorate of the European Commission's Directorate General for Research and Innovation launched a consultation through a questionnaire of all participants involved in health research projects from 2002 to 2010. The EU funding over this period. for projects involving collaboration between research teams, reached around (sic)4.9 billion and involved 12,460 teams in I 14 different countries. This survey was performed to contribute to the preparation the next Framework Programme for research and innovation called "Horizon 2020" (2014-2020). A statistical analysis of the questionnaire replies has been carried out. This paper provides a short summary of the most salient features and results of this survey. The survey reveals a significant scientific and socio-economic impact of EU funding in the area of health research and a positive long-term impact of this funding on health research capacities in Europe.

International audience; In this paper, an analysis of the coronary trees using Magnetic Resonance Angiography (MRA) is performed. The objective is to estimate how much MRA is capable to provide insights into the vascular network. A qualitative exploration of the MRA volumes with anatomical labelling by experts is rst performed, Quantitative vessel features are then manually extracted providing a ground truth which is further compared to a semi-automatic extraction. This evaluation is carried out on 10 datasets of the SSFP MRA sequence and allows getting a more precise view on the current state-of-the- art as well as on future achievements to be done.

Shape signatures are fundamental in medicine to characterize abnormal patterns and track their evolution over time. Beyond the diagnosis objective, they are critical to understand the relations between structures and functions as well as to design advanced image-guided therapy. One of the most critical steps in medical imaging remains segmentation, aimed at a fast, precise, robust and reproducible extraction of the organ boundaries. We present a 3D semi-automatic technique for segmenting the inner boundary of abdominal aortic aneurysms (AAA). The method that is proposed makes use of recent developments from graph theory, the graph-cut algorithms. It requires a limited interaction with the user (for learning the object and background statistical features). Its performance is highlighted on multi-detector CT angiography data. Some perspectives are then sketched in order to show how this result could be improved to feed fluid-structure simulations.

The estimation of coronary vessel motion from a sequence of digital subtraction angiographic (DSA) images is presented. It is carried out by means of a gradient-based method along the arterial centerlines from two successive images. To extract the centerlines from all the DSA images of a sequence, an automatic, model-guided, frame-to-frame feature extraction approach is used, in which the estimated motion velocities available in a previous image are included to initiate the model. The overall process has been implemented and tested using several sequences. The results are presented. >

Tridimensional (3D) visualization of coronary arteries offers an excellent way to improve the diagnostic and the therapeutical evaluation of certain pathological coronary affections. Reconstruction of 3D images based on two bidimensional projections (Image 2D) is unfortunately a multiple solution problem. Identification of the correct solution is only possible using a priori knowledge, which includes 2D and 3D aspects. In this work, using visual resources, the authors study the influence of 3D proprieties in reducing the search space (set of possible solutions). Particularly the ventricular surface generated using the coronary "arteries. Research carried out using real data (P. Windyga, Doctoral Thesis, 1994) shows that the 3D properties useful in identifying the correct solution are: arteries not crossing each other, arteries not crossing the ventricular cavity and all the arteries wrapping an elipsoidal space. The modelling used shows that with the application of these properties the authors can reduce the search space by at least 30%. Nevertheless it was realized that properties of the generated ventricular surface are sensible to the choice of the reference ventricular axis. The results obtained are highly promising and encourage further research.

The recent advances in medical imaging, the avaibility of methods for image analysis and computer graphics, the technological resources provided by microdevices and the design of minimal access surgical procedures have open the road to new concepts. Virtual endoscopy represents one of these emerging areas and points out the applicative potential of three dimensional (3D) imaging. It leads to less invasive diagnosis and therapeutic achievements and provides important cues for education and interventional planning. Image segmentation, visualization, tissue modeling and interactions with surgical instruments are the fundamental components to build clinical applications. They are examined in this paper through 3D navigation systems, surgical simulations and image guided interventions.

In the field of epilepsy, the analysis of stereoelectroencephalographic (SEEG) signals recorded with depth electrodes provides major information on interactions between brain structures during seizures. A comprehensive methodology of comparing SEEG seizure recordings is presented. It proceeds in three steps: 1) segmentation of SEEG signals; 2) characterization and labeling of segments; and 3) comparison of observations coded as sequences of symbol vectors. The third step reports a vectorial extension of the Wagner and Fischer's algorithm to first, quantify similarities between observations and second, extract invariant sequences of events, referred to as spatio-temporal signatures. The study shows that two observations of nonequal duration can be matched by deforming the first one to optimally fit the second, under cost constraints. Results show that the methodology allows to exhibit signatures occurring during epileptic seizures and to point out different types of seizure patterns. The study brings objective results on reproducible interactions between brain structures during ictal periods and may help in the understanding of epileptogenic networks.

This paper presents a review of the applications of texture analysis in medical imaging. Many authors take a great interest in this topic (75 papers have been published since 1984) and try to elaborate automatic methods for tissue characterization. The results are not really convincing and applications are often reduced to feasibility studies. This failure is due to the empirical approach to the problem: the first studies were performed on ultrasound images, in which visual texture is very present, but no data standardization is available with this imaging modality. A more rational approach should provide better results. For each organ or tissue, it is necessary to find the appropriate source and texture analysis method. This difficult task requires reflection concerning the interactions between tissues and imaging sources, to define judicious structuring elements. These structuring elements should facilitate the choice of the best texture analysis method, for the particular application. Considerable methodological progress has yet to be made, after which texture analysis should be a useful and efficient tool for clinical use.

The objective of this article is to define the different stages involved in the 3D reconstruction of arteries and to review, from our experience and from the literature, the solutions already proposed. A full reconstruction framework includes the characterization of the imaging device (in terms of distortion and calibration), the specificity of the image acquisition process, the preprocessing that can be applied, the detection of the vascular structures, the 2D feature formation, the reconstruction itself, and the visualization aspects. They are examined according to a computer vision approach where two or three views are assumed to be available. Their generalization to temporal image sequences are also considered. Some of the material reported here is unpublished. The article allows the reader to identify the true critical issues that are not often clearly mentioned in the literature and the challenges that they convey. A final discussion presents a few perspectives in this area of research.

A framework for coronary vessels analysis in digital subtracted angiograms is described. This method combines the motion estimation with the frame-to-frame structure detection in a natural way such that they act interactively. The first step consists of the extraction of the vessel centrelines in one image and their organization into meaningful constituents or branches of the coronary arterial tree. The motion is then estimated along the centrelines through a gradient based method. These motion estimates supply an initial positioning of an active contour model (or ‘snake“) in the next image. This model adapts itself by changing its shape to accurately fit onto the new centrelines. This process is then reiterated on the subsequent images to depict the dynamic behaviour of all the relevant branches. The main interests of this scheme are: (1) the active models operate locally so a fast detection of the vessels can be performed; (2) the centrelines extraction is fully guided by the confluence of the motion estimation and the contour model; (3) both morphological and kinetic features are provided on a quantitative basis.

An automatic and quantitative framework for the analysis of surface EMG interference patterns at maximal non-controlled force is described. The main contents are focused on different ways to represent surface EMG signals and to analyse collected data set through principal component analysis. The methods are tested on control individuals and on patients with Guillain-Barré Syndrome examined at different moments after the beginning of the disease (first week, three months, and one year). The results exhibit that a high diagnostic yield can be provided and furthermore that quantification and differentiation can be expected. Surface EMG interference pattern may be hoped to be a useful tool in daily practice.

KISS is the acronym for Knowledge-based Intelligent Signal monitoring System, a European Communities project concerning the application of knowledge-based systems in medical monitoring systems.

Emphasis has been recently given to automatic segmentation of vascular network observed by means of Digital Subtraction Angiography. This paper details a new algorithm which determines the medial axes and the contours of the vessesl in the whole vascular tree.

The real-time and automatic system that we have developed, providing EMG signal acquisition, processing and storage, helps in decision making. This implies the formation of a reference signal set and the accurate determining of the influence of various parameters. Using data analysis methods, our normal population analysis investigates the effects of age, sex and muscle. Assignment tests have been carried out with learning classes (formed with signals obtained from healthy and myopathic subjects) to measure the variability of a possible aid-to-diagnosis procedure.

An integrated interactive software system called 'X-Image 3D' has been developed for the display and the analysis of 3D and multidimensional images. The authors show how new functionalities can be designed to deal with highly complex and data intensive applications such as the understanding of epileptic patient diseases. >

Resume Une approche originale du traitement de l'information electromyographique est presentee, dont le principal objectif est le diagnostic des pathologies neuro-musculaires. La structure du systeme est basee sur trois modules (SIGMA, DIALOG, EXPERT) chacun possedant une base de connaissances locale specialisee. Les principales caracteristiques sont presentees: son originalite tient, d'une part, a sa structure modulaire, distribuee et hierarchisee, et, d'autre part, a sa coherence conceptuelle permettant d'integrer des methodologies de traitement du signal, d'analyses de donnees et d'intelligence artificielle. Les avantages de cette approche se mesureront en terme d'optimisation de l'examen electromyographique.