Search

International audience; Deep neural networks have recently surpassed other image restoration methods which rely on hand-crafted priors. However, such networks usually require large databases and need to be retrained for each new modality. In this paper, we show that we can reach nearoptimal performances by training them on a synthetic dataset made of realizations of a dead leaves model, both for image denoising and superresolution. The simplicity of this model makes it possible to create large databases with only a few parameters. We also show that training a network with a mix of natural and synthetic images does not affect results on natural images while improving the results on dead leaves images, which are classically used for evaluating the preservation of textures. We thoroughly describe the image model and its implementation, before giving experimental results.

International audience; In this chapter, we evaluate the 3D tumor trajectories from patient-specific biomechanical model of the respiratory system, which takes into account the physiology of respiratory motion to simulate irregular motion. The behaviour of the lungs, driving directly by simulated actions of the breathing muscles, i.e. the diaphragm and the intercostal muscles (the rib cage). In this chapter, the lung model is monitored and controlled by a personalized lung pressure-volume relationship during a whole respiratory cycle. The lung pressure is patient specific and calculated by an optimization framework based on inverse finite element analysis. We have evaluated the motion estimation accuracy on two selected patients, with small and large breathing amplitudes (Patient 1 = 10.9 mm, Patient 10 = 26.06 mm). In this order, the lung tumor trajectories identified from 4D CT scan images were used as reference and compared with the 3D lung tumor trajectories estimated from finite element simulation during the whole cycle of breathing. Over all phases of respiration, the average mean error is less than 1.8 ± 1.3 mm. We believe that this model, despite of others takes into account the challenging problem of the respiratory variabilities and can potentially be incorporated effectively in Treatment Planning System (TPS) and as lung tumor motion tracking system during radiation treatment.

Transfer learning from huge natural image datasets, fine-tuning of deep neural networks and the use of the corresponding pre-trained networks have become de facto the core of art analysis applications. Nevertheless, the effects of transfer learning are still poorly understood. In this paper, we first use techniques for visualizing the network internal representations in order to provide clues to the understanding of what the network has learned on artistic images. Then, we provide a quantitative analysis of the changes introduced by the learning process thanks to metrics in both the feature and parameter spaces, as well as metrics computed on the set of maximal activation images. These analyses are performed on several variations of the transfer learning procedure. In particular, we observed that the network could specialize some pre-trained filters to the new image modality and also that higher layers tend to concentrate classes. Finally, we have shown that a double fine-tuning involving a medium-size artistic dataset can improve the classification on smaller datasets, even when the task changes., Accepted at Workshop on Fine Art Pattern Extraction and Recognition (FAPER), ICPR, 2020

International audience; While pap test is the most common diagnosis methods for cervical cancer, their results are highly dependent on the ability of the cytotechnicians to detect abnormal cells on the smears using brightfield microscopy. In this paper, we propose an explainable region classifier in whole slide images that could be used by cyto-pathologists to handle efficiently these big images (100,000x100,000 pixels). We create a dataset that simulates pap smears regions and uses a loss, we call classification under regression constraint, to train an efficient region classifier (about 66.8% accuracy on severity classification, 95.2% accuracy on normal/abnormal classification and 0.870 KAPPA score). We explain how we benefit from this loss to obtain a model focused on sensitivity and, then, we show that it can be used to perform weakly supervised localization (accuracy of 80.4%) of the cell that is mostly responsible for the malignancy of regions of whole slide images. We extend our method to perform a more general detection of abnormal cells (66.1% accuracy) and ensure that at least one abnormal cell will be detected if malignancy is present. Finally, we experiment our solution on a small real clinical slide dataset, highlighting the relevance of our proposed solution, adapting it to be as easily integrated in a pathology laboratory workflow as possible, and extending it to make a slide-level prediction.

International audience; Deep learning methods are widely used for medical applications to assist medical doctors in their daily routine. While performances reach expert's level, interpretability (highlighting how and what a trained model learned and why it makes a specific decision) is the next important challenge that deep learning methods need to answer to be fully integrated in the medical field. In this paper, we address the question of interpretability in the context of whole slide images (WSI) classification with the formalization of the design of WSI classification architectures and propose a piece-wise interpretability approach, relying on gradient-based methods, feature visualization and multiple instance learning context. After training two WSI classification architectures on Camelyon-16 WSI dataset, highlighting discriminative features learned, and validating our approach with pathologists, we propose a novel manner of computing interpretability slide-level heat-maps, based on the extracted features, that improves tile-level classification performances. We measure the improvement using the tile-level AUC that we called Localization AUC, and show an improvement of more than 0.2. We also validate our results with a RemOve And Retrain (ROAR) measure. Then, after studying the impact of the number of features used for heat-map computation, we propose a corrective approach, relying on activation colocalization of selected features, that improves the performances and the stability of our proposed method.

For the development of a water quality modeling classification, parameter optimization is important. In this research, in order to enhance the strength of the used approach, we propose a hybrid approach that combines SVM classifiers with PSO and PCA selection features. This is used for classifying the status of water quality with the Radial Basis Function (RBF) SVM kernel. To enhance the classification accuracy, PSO selects the best parameter for SVM. The problem of irrelevant data in the space of functions can be solved by PCA. A binary classification based on two water quality classes (Class I: upper, Class II: lower) is considered to be the problem. Datasets were obtained for training and testing over 5 years (2014-2018) from many samples in Tilsdit dam-Algeria, and are used in this situation. A simulation of the training time and recognition rate will be carried out in order to verify the efficiency of the method. The results obtained demonstrate that the proposed method had great potential for classifying water quality.

One of the key causes of premature disability and mortality in the world today is heart disease, which makes its prediction a vital problem in the field of healthcare systems. This work provides a contribution to the study and creation an intelligent system based on LSTM technique for heart disease prediction. A comparative study is presented between Multi Layer Perceptron (MLP) and Long Short Term Memory (LSTM) techniques in terms of accuracy and other predictive parameters for heart disease. The main aim is to develop an intelligent system based on LSTM technique for predicting heart disease in order to make an adapted decision to prevent and monitor heart disease and stroke. As it has better characteristics than those of the MLP technique, LSTM is shown to be the most effective technique for solving the aforementioned problems.

International audience; In this paper, we have applied and compared two scenarios: 1) The linear pressure (ramp amplitude) for the whole cyclewithout any optimization of the lung pressure at different respiratory states (named without optimization) and 2) Optimized lung pressure and diaphragm force using an automatic tuning algorithm based on inverse FE analysis methodology during a whole respiratory cycle and at different states (with optimization).

Weakly supervised object detection (WSOD) using only image-level annotations has attracted a growing attention over the past few years. Whereas such task is typically addressed with a domain-specific solution focused on natural images, we show that a simple multiple instance approach applied on pre-trained deep features yields excellent performances on non-photographic datasets, possibly including new classes. The approach does not include any fine-tuning or cross-domain learning and is therefore efficient and possibly applicable to arbitrary datasets and classes. We investigate several flavors of the proposed approach, some including multi-layers perceptron and polyhedral classifiers. Despite its simplicity, our method shows competitive results on a range of publicly available datasets, including paintings (People-Art, IconArt), watercolors, cliparts and comics and allows to quickly learn unseen visual categories., Comment: 26 pages, 12 figures

Deep learning methods are widely used for medical applications to assist medical doctors in their daily routines. While performances reach expert's level, interpretability (highlight how and what a trained model learned and why it makes a specific decision) is the next important challenge that deep learning methods need to answer to be fully integrated in the medical field. In this paper, we address the question of interpretability in the context of whole slide images (WSI) classification. We formalize the design of WSI classification architectures and propose a piece-wise interpretability approach, relying on gradient-based methods, feature visualization and multiple instance learning context. We aim at explaining how the decision is made based on tile level scoring, how these tile scores are decided and which features are used and relevant for the task. After training two WSI classification architectures on Camelyon-16 WSI dataset, highlighting discriminative features learned, and validating our approach with pathologists, we propose a novel manner of computing interpretability slide-level heat-maps, based on the extracted features, that improves tile-level classification performances by more than 29% for AUC., Comment: 8 pages (references excluded), 3 figures, presented in iMIMIC Workshop at MICCAI 2020

Water quality monitoring plays a vital role in the protection of water resources, environmental management, and decision-making. Artificial intelligence (AI) based on machine learning techniques has been widely used to evaluate and classify water quality for the last two decades. However, traditional machine learning techniques face many limitations, the most important of which is the inability to apply these techniques with big data generated by smart water quality monitoring stations to improve the prediction. Real-time water quality monitoring with high accuracy and efficiency for intelligent water quality monitoring stations requires new and sophisticated techniques based on machine and deep learning techniques. For this purpose, we propose a novel approach based on the integration of deep learning and feature extraction techniques to improve water quality classification. In this paper, was chosen the Tilesdit dam in Bouira (Algeria) as a case study. Moreover, we implemented the advanced deep learning method - Long Short Term Memory Recurrent Neural Networks (LSTM RNNs) to construct an intelligent model for drinking water quality classification. Furthermore, principal component analysis (PCA), linear discriminant analysis (LDA) and independent component analysis (ICA) techniques were used for features extraction and data reduction from original features. Additionally, we used three methods of cross-validation and two methods of the out-of-sample test to estimate the performance of LSTM RNNs model. From the results we found that the integration of LSTM RNNs with LDA, and LSTM RNNs with ICA yields an accuracy of 99.72%, using Random-Holdout technique.

This is an extended version of HAL preprint hal-01676326; International audience; Image synthesis is a core problem in modern deep learning, and many recent architectures such as autoencoders and Generative Adversarial networks produce spectacular results on highly complex data, such as images of faces or landscapes. While these results open up a wide range of new, advanced synthesis applications, there is also a severe lack of theoretical understanding of how these networks work. This results in a wide range of practical problems, such as difficulties in training, the tendency to sample images with little or no variability, and generalisation problems. In this paper, we propose to analyse the ability of the simplest generative network, the autoencoder, to encode and decode two simple geometric attributes : size and position. We believe that, in order to understand more complicated tasks, it is necessary to first understand how these networks process simple attributes. For the first property, we analyse the case of images of centred disks with variable radii. We explain how the autoencoder projects these images to and from a latent space of smallest possible dimension, a scalar. In particular, we describe a closed-form solution to the decoding training problem in a network without biases, and show that during training, the network indeed finds this solution. We then investigate the best regularisation approaches which yield networks that generalise well. For the second property, position, we look at the encoding and decoding of Dirac delta functions, also known as `one-hot' vectors. We describe a hand-crafted filter that achieves encoding perfectly, and show that the network naturally finds this filter during training. We also show experimentally that the decoding can be achieved if the dataset is sampled in an appropriate manner.

In order to overcome the limited depth of field of usual photographic devices, a common approach is multi-focus image fusion (MFIF). From a stack of images acquired with different focus settings, these methods aim at fusing the content of the images of the stack to produce a final image that is sharp everywhere. Such methods can be very efficient, but when a global geometric alignment of images is out-of-reach, or when some objects are moving, the final image shows ghosts or other artefacts. In this paper, we propose a generic method to overcome these limitations. We first select a reference image, and then, for each image of the stack, reconstruct an image that shares the geometry of the reference and the sharpness content of the image at hand. The reconstruction is achieved thanks to a specially crafted modification of the PatchMatch algorithm, adapted to blurred images, and to a dedicated postprocessing for correcting reconstruction errors. Then, from the new image stack, MFIF is performed to produce a sharp result. We show the efficiency of the result on a database of challenging cases of hand-held shots containing moving objects.

International audience; Respiratory-induced organ motion is a technical challenge to PET imaging.This motion induces displacements and deformation of the organs tissues, which need to be taken into account when reconstructing the spatial radiation activity. Classical image-based methods that describe motion using Deformable Image Registration (DIR) algorithms cannot fully take into account the non-reproducibility of the respiratory internal organ motion nor the tissue volume variations that occur during breathing. In order to overcome these limitations, various biomechanical models of the respiratory system have been developed in the past decade as an alternative to DIRapproaches. In this paper, we describe a new method of correcting motion artefacts in PET imagereconstruction adapted to motion estimation models such as those based onthe finite element method (FEM). In contrast with the DIR-based approaches, the radiation activity was reconstructed on deforming tetrahedral meshes. For this, we have re- formulated the tomographic reconstruction problem by introducing atime-dependent system matrix based calculated using tetrahedral meshes instead of voxelized images.The MLEM algorithm was chosen as the reconstruction method. The simulationsperformed in this study show that the motion compensated reconstruction based on tetrahedral deformable meshes has the capability to correct motion artefacts. Results demonstrate that, in the case of complex deformations, when large volume variations occur, the developed tetrahedral based method is more appropriate than the classical DIR-based one. This method can be used, together with biomechanical models controlled by external surrogates, to correct motion artefacts in PET images and thus reducing the need for additional internal imaging during the acquisition.

International audience; Respiratory-induced organ motion is a technical challenge tonuclear imaging and to particle therapy dose calculations forlung cancer treatment in particular. Internal organ tissue dis-placements and deformations induced by breathing need to betaken into account when calculating Monte Carlo dose distri-butions or when performing tomographic reconstructions forPET imaging. This paper proposes a method to reconstructPET activities over tetrahedral meshes which are deformedbased on biomechanical patient specific model of the respi-ratory system to tackle the non reproductibility of the breath-ing. We also describe the adaptation of the popular List-ModeMaximum Likelihood Estimation (LM-MLEM) reconstruc-tion algorithm to motion estimation model using the finite el-ement method (FEM). Our simulations demonstrate the accu-racy of the proposed 4D LM-MLEM reconstruction algorithmbased on biomechanical model and its capability to correctmotion artifacts due to the breathing.

Synthetic aperture radar (SAR) images display very high dynamic ranges. Man-made structures (like buildings or power towers) produce echoes that are several orders of magnitude stronger than echoes from diffusing areas (vegetated areas) or from smooth surfaces (e.g., roads). The impulse response of the SAR imaging system is, thus, clearly visible around the strongest targets: sidelobes spread over several pixels, masking the much weaker echoes from the background. To reduce the sidelobes of the impulse response, images are generally spectrally apodized, trading resolution for a reduction of the sidelobes. This apodization procedure (global or shift-variant) introduces spatial correlations in the speckle-dominated areas that complicates the design of estimation methods. This paper describes strategies to cancel sidelobes around point-like targets while preserving the spatial resolution and the statistics of speckle-dominated areas. An irregular sampling grid is built to compensate the subpixel shifts and turn cardinal sines into discrete Diracs. A statistically grounded approach for point-like target extraction is also introduced, thereby providing a decomposition of a single look complex image into two components: a speckle-dominated image and the point-like targets. This decomposition can be exploited to produce images with improved quality (full resolution and suppressed sidelobes) suitable both for visual inspection and further processing (multitemporal analysis, despeckling, interferometry).

Biomechanical patient-specific computational modelling and simulation of the respiratory system is one of the important areas of research in medical imaging and radiation therapy. Lung tumor respir...

International audience; Purpose: This paper investigates the capabilities of a dual-rotation C-arm cone-beam computed tomography (CBCT) framework to improve non-contrast-enhanced low-contrast detection for full volume or volume-of-interest (VOI) brain imaging.Method: The idea is to associate two C-arm short-scan rotational acquisitions (spins): one over the full detector field of view (FOV) at low dose, and one collimated to deliver a higher dose to the central densest parts of the head. The angular sampling performed by each spin is allowed to vary in terms of number of views and angular positions. Collimated data is truncated and does not contain measurement of the incoming X-ray intensities in air (air calibration). When targeting full volume reconstruction, the method is intended to act as a virtual bow-tie. When targeting VOI imaging, the method is intended to provide the minimum full detector FOV data that sufficiently corrects for truncation artifacts. A single dedicated iterative algorithm is described that handles all proposed sampling configurations despite truncation and absence of air calibration.Results: Full volume reconstruction of dual-rotation simulations and phantom acquisitions are shown to have increased low-contrast detection for less dose, with respect to a single-rotation acquisition. High CNR values were obtained on 1% inserts of the Catphanmath formula 515 module in 0.94 mm thick slices. Image quality for VOI imaging was preserved from truncation artifacts even with less than 10 non-truncated views, without using the sparsity a priori common to such context.Conclusion: A flexible dual-rotation acquisition and reconstruction framework is proposed that has the potential to improve low-contrast detection in clinical C-arm brain soft-tissue imaging.

International audience; The aim of this work is to present a resampling scheme for SAR images that preserves spatial resolution and produces statistically accurate images at the same time. Indeed, SAR images are, for reasons due to their acquisition process, well sampled signals according to the Shannon sampling theory. In the presence of strong responses, that we will refer to as targets, a sinc-like function centered at the target is smeared over the entire image and is particularly visible in the range of tens of pixels surrounding the target. To mitigate this phenomenon, the usual solution is to apply an apodization window in the Fourier domain so as to change the cardinal sine impulse response into a much rapidly decaying one. This approach has two major drawbacks. It reduces the resolution of the image and introduces inaccurate statistical dependency between pixels. We propose to resample the image in an adaptive and robust way so that the target smear is canceled and the new sampled image is completely faithful to the underlying signal.

International audience; Considered as the second men cancer in Algeria, prostate cancer is treated in 70% by radiation. That's why radiation therapy is therapeutic weapon for prostate cancer. Conformational Radiotherapy in 3D is the most common technique [1−5]. The use of conventionally optimized treatment plans was compared at case scenario of optimized treatment plans VMAT for prostate cancer. The evaluation of the two optimizations strategies focused on the resulting plans ability to retain dose objectives under the influence of patient set up. Dose Volume Histogram in the Planning Target Volume and dose in the Organs At Risks were used to calculate the conformity index, and evaluation ratio of irradiated volume which represent the main tool of comparison [6,7]. The situation was analysed systematically. The 14% dose increase in the target leads to a decrease in the dose in adjacent organs with 39% in the bladder. Therefore, the criterion for better efficacy and less toxicity reveal that VMAT is the best choice.

Purpose: The estimation of the distribution patterns of energy and dose in respiratory-induced organ motion represents a technical challenge for hadron therapy treatment planning, notably in the case of lung cancer in which many difficulties arose, like tissue densities variation and the tumor position shifting during breathing. This study focuses on the comparison between deformable tetrahedral meshes and voxel-based structures used as computational phantoms in four-dimensional dose calculations. The former use a continuous representation of tissue densities by respecting mass conservation principle, while the latter is a discrete grid of density values (CT-scan). Methods: The movement used to simulate breathing is generated with deformable image registration (DIR) of CT images (Castillo, 2010) (Klein, 2010) (Shamonin, 2013). Tissue tracking for tetrahedral model is implicitly performed by the fact that the meshes maintain their topology during deformations. The dose distribution is calculated using the time-dependent tetrahedral density map issued from 4D-CT scans (Petru Manescu, 2014). Unlike image-based methods, the deposited energy is accumulated inside each deforming tetrahedron of the meshes. An implementation of this dose computation method on a deformable anatomy in the case of a passive scattering beam line is demonstrated using the Geant4 code (Agostinelli, 2003). Besides, energy values in voxel-based structures are calculated for each time step and accumulated using the transformations provided by the registration. Then, values are accumulated back onto the reference image and divided by the mass to obtain the 4D dose map. Figure 1 illustrates the process used to accumulate dose in respiratory-induced simulations. Results: The tetrahedral mesh dose distribution was compared to the conventional voxel-based structure using a thoracic 4D-CT data of a patient case. Preliminary results show that dose distributions for both representations are in a good agreement (figure 2), and dose homogeneity is about the same (table1). However, motion-induced dose accumulations are more intuitive using a tetrahedral model since they do not introduce additional uncertainties with image resampling and interpolation methods, and also for the fact that they respect mass conservation principle. Conclusion: We have developed a 4D tetrahedral model for Monte Carlo dose calculations alongside its implementation on the Geant4 platform. Results of comparison with conventional methods based on voxels have shown that dose distributions are in good agreement. This novel structure can be of a great aid for treatment planning of moving targets. An experimental validation based on 4D anthropomorphic phantom (e.g. LuCa phantom developed in paul scherrer institute) (Neihart, 2013) would draw a clear conclusion regarding the performance of the presented method in comparison with the classical methods. Nevertheless, the main advantage of this method is that, coupled with a patient-specific biomechanical model, it could be used in the future to correct motion artefacts in treatment planning.

Organ motion due to patient breathing introduces a technical challenge for dosimetry and lung tumor treatment by hadron therapy. Accurate dose distribution estimation requires patient-specific information on tumor position, size, and shape as well as information regarding the material density and stopping power of the media along the beam path. A new 4D dosimetry method was developed, which can be coupled to any motion estimation method. As an illustration, the new method was implemented and tested with a biomechanical model and clinical data. First, an anatomical model of the lung and tumor was synthesized with deformable tetrahedral grids using computed tomography (CT) images. The CT attenuation values were estimated at the grid vertices. Respiratory motion was simulated biomechanically based on nonlinear finite element analysis. Contrary to classical image-based methods where motion is described using deformable image registration algorithms, the dose distribution was accumulated over tetrahedral meshes that are deformed using biomechanical modeling based on finite element analysis. The new method preserves the mass of the objects during simulation with an error between 1.6 and 3.6 %. The new method was compared to an existing dose calculation method demonstrating significant differences between the two approaches and overall superior performance using the new method. A unified model of 4D radiotherapy respiratory effects was developed where biomechanical simulations are coupled with dose calculations. Promising results demonstrate that this approach has significant potential for the treatment for moving tumors.

Water quality can be difficult to measure. Each water body can contain dramatically different levels of pollution. Water quality issues influence human and environmental health, so the more we monitor our water the better we will be able to recognize and prevent contamination problems. This paper presents the performance evaluation of Artificial Neural Network (ANN) and Support Vector Machines (SVM) classification techniques using Dempster-Shafer Theory of Evidence (DSTE) combination data fusion in monitoring of water quality. Dempster-Shafer Theory of Evidence (DSTE) was applied to increase the classification accuracy. This study involved the evaluation and interpretation of surface water quality data in Tilesdit dam (Algeria). It also allowed us to obtain more advanced information about water quality, and to design a monitoring network for this study area. The ANN and SVM which are two techniques for pattern classification have been widely used in many application areas such as water quality monitoring. These methods are binary classification techniques, but in some cases, such as pattern recognition, we need more than two classes. One of the suggested solutions for this difficulty is to split the problem into a set of binary classification before combining them. A multi-class problem using ANN and SVM is a typical example for solving the mentioned problem. The MLP network and the algorithm of SVM, one-against-all, are the most popular strategies for multi-class problems. In order to manage the conflict that results from using both approaches, the final decision is performed using DSTE's rule combination based on probabilistic output from both the two classifiers. In this work, four physicochemical parameters in 4 seasons during the period 2009–2011, located at Tilesdit dam in Bouira (Algeria), were selected for the purpose of this study, such as temperature, pH, conductivity and turbidity to supervise water quality. In order to evaluate their performances, a simulation using real dataset, corresponding to the recognition rate (training and test), is carried out. The results are compared to get the best performance evaluation of classification process which depends strongly on the choice of the excellent characteristics in training phase. The data fusion method improved significantly the performance of the intelligent approach in water quality classification. Moreover, the results demonstrated that the proposed procedure had great potential in water quality monitoring.

Control of the helicopter includes nonlinearities, uncertainties and external perturbations that should be considered in the design of control laws. This paper presents a control strategy for TRMS (twin rotor mimo system), based on the sliding mode control using non linear sliding surfaces. The choice of this algorithm was justified by his robustness with respect to modeling errors and external disturbances. The simulation results of the proposed algorithm control show good performances. Finally simulation results are included to indicate the TRMS system with the proposed controller can greatly alleviate the chattering effect and remain a good trajectory tracking.

The distribution patterns estimation of energy and dose in respiratory-induced organ motion represents a technical challenge for hadron therapy treatment planning. This paper focuses on the comparison between deformable tetrahedral meshes and the conventional voxel-based structures, used as computational phantoms in four-dimensional dose calculations. The movement issued from deformable image registration of CT-images is used for both representations to calculate the 4D dose distributions. An implementation of the dose computation method for tetrahedral meshes on a deformable anatomy in case of passive scattering beam line is given using the Geant4 code. Tissue tracking is implicitly performed by the fact that the meshes maintain their topology during deformations, and the dose distribution is calculated using time-dependent tetrahedral density map issued from 4D-CT scans. Our results show that dose distributions for voxel and tetrahedral meshes are in good agreement, and dose homogeneity is almost the same.

Images acquired by the Moderate Resolution Imaging Spectroradiometer (MODIS) aboard Terra and Aqua exhibit strong detector striping. This artifact is common to most pushbroom scanners and affects both visual interpretation and radiometric integrity of remotely sensed data. A considerable effort has been made to remove stripe noise and reduce its impact on high-level products. Despite the variety of destriping algorithms proposed in the literature, complete removal of stripes without signal distortion is yet to be overcome. In this paper, we tackle the striping issue from a variational angle. Basic statistical assumptions used in previous techniques are replaced by a much realistic geometrical consideration on the striping unidirectional variations. The resulting algorithm is tested on Aqua and Terra MODIS data contaminated with severe stripes and is shown to provide optimal qualitative and quantitative results.

This paper proposes a truly interactive virtual environment (VE) system for 2-D assembly tasks at the microscale. It is based on the application of virtual potential fields as a control aid for performing safe and reliable path planning strategies. The planner covers a whole range of problems due to microscale effects in object assignment, obstacle detection and avoidance, path trajectory finding and sequencing. We investigated various paradigms for enabling the human operator and the automatic motion planner to cooperatively solve a motion planning task through the use of virtual potential fields. Communication between the operator and the planner is made through haptic/vision/sound modalities. First, we describe algorithms based on optimization theory and Voronoi graph construction taking into account the microscale effects. As automatic motion planners fail due to the difficulty of discovering critical configurations, we propose cooperation paradigms with operator skills in order to solve motion planning strategies. Then, potential fields are being used as a tool to generate velocity commands from an automatic path planner as well as allowing the human to interact. Finally, the ideas presented here are supported by experiments for efficient pushing-based manipulation constructing 2-D microparticle patterns.

International audience; Sparse modeling can be used to characterize outlier type noise. Thanks to sparse recovery theory, it was shown that 1-norm super-resolution is robust to outliers if enough images are captured. Moreover, sparse modeling of signals is a way to overcome ill-posedness of under-determined problems. This naturally leads to the question: does an added sparsity assumption on the signal will improve the robustness to outliers of the 1-norm super-resolution, and if yes, how strong should this assumption be? In this article, we review and extend results of the literature to the robustness to outliers of overdetermined signal recovery problems under sparse regularization, with a convex variational formulation. We then apply them to general random matrices, and show how the regularization parameter acts on the robustness to outliers. Finally, we show that in the case of multi-image processing, the structure of the support of signal and noise must be studied precisely. We show that the sparsity assumption improves robustness if outliers do not overlap with signal jumps, and determine how the regularization parameter can be chosen.

International audience; Patient-specific respiratory motion modeling may help to understand pathophysiology and predict therapy planning. The respiratory motion modifies the shape and position of internal organs. This may degrade the quality of such medical acts as radiotherapy or laparoscopy. Predicting the breathing movement is complex, and it is considered as one of the most challenging areas of medical research. This paper presents a biomechanical model of the respiratory system, based on the finite element method (FEM), including the biomechanical behavior of the diaphragm as well as rib kinematics computations, on the assumption that breathing is controlled by two independent actors: the thorax and diaphragm muscles. In order to predict the type of the (geometrical or material) nonlinearities, a quantitative comparison of the clinical data was applied on 12 patients. We propose two nonlinear hyperelastic models: the Saint-Venant Kirchhoff and Mooney–Rivlin models. Our results demonstrate that the nonlinear hyperelastic Mooney–Rivlin model of the diaphragm behaves similarly to the linear elastic model with large displacement (Saint-Venant Kirchhoff). The results suggest that the approach of small strains (within the large displacement) may be globally maintained in the modeling of the diaphragm, and demonstrate that the accuracy of the proposed FEM is capable to predict the respiratory motion with an average surface error in a diaphragm/lungs region of interest contact of 2. 0 ± 2. 3 mm for the contact surface between lungs and diaphragm. The comparison study between the FEM simulations and the CT scan images demonstrates the effectiveness of our physics-based model.

International audience; Organ motion, especially respiratory motion, is a technical challenge to hadrontherapy planning and dosimetry. This motion induces the displacement and the deformation of the organs tissues along the radiation beam path which need to be taken into account when computating dose distribution during the treatment. In this paper we present an original approach of virtual patient modeling for 4D radiation therapy simulations. As opposed to classical image-based models, where the necessary information is distributed over a rigid structured grid of voxels, we represent the human anatomy with a deformable grid of tetrahedra where the mass density is mapped to the vertices of the grid. In this way, we can simulate within the same structure organ motion, mass density variations and dose distribution without having to perform voxel tissue tracking.

International audience; Microinjection is a method for the delivery of exogenous materials into cells and is widely used in biomedical research areas such as transgenics and genomics. However, this direct injection is a time-consuming and laborious task, resulting in low throughput and poor reproducibility. Here, we describe a telerobotic shared control framework for microinjection, in which a micromanipulator is controlled by the shared motion commands of both the human operator and the autonomous controller. To determine the weightings between the operator and the controller, we proposed a quantitative evaluation method using a model of speed/accuracy trade-offs in human movement. The results showed that a 40%-60% weighting on the human operator (or the controller) produced the best performance for both speed and accuracy of guiding and targeting task in microinjection suggesting that some level of both automation and human involvement is important for microinjection tasks.

International audience; This paper presents a complete computer-based biological training system to simulate intracytoplasmic sperm injection (ICSI) procedures in a virtual environment (VE). Using the proposed system, the technician can practice and improve his control by developing a gesture similar to that performed in reality. The human-machine user's interface also provides an efficient haptic-based guidance tool to minimize the risk of damaging membranes during the injection. The key challenge of deformable simulation is to satisfy the conflicting requirements of real-time interactivity and physical realism. The compromise between the realism of bio-mechanical models used, the accuracy and stability of computational algorithms for real-time haptic rendering are fullfilled with the use of a mechanical modeling based on a hyperelastic model (St-Venant-Kirchhoff) and the writing of a specific dynamic finite element code. Simulated responses are compared to experimental data to show the effectiveness of the proposed physically-based model.

International audience; Nowadays, a key challenge of deformable simulation is to satisfy the conflicting requirements of real-time interactivity and physical realism. Various models have been implemented, the soft tissue modeling are usually derived using either mass-spring approach or the linear finite element method (FEM). The linear elasticity is often used for the modeling of deformable materials, mainly because the equations remain quite simple and the computation time can be optimized. The physical behavior of soft tissue may be considered as linear elastic if its displacement and deformation remain small (typically less than 10 % of the mesh size). The objective of this paper is to develop and implement an interactive simulation techniques to facilitate training of biological cell injection operations. Using this tool, the operator can form, train and improve its control by developing a gesture similar to that performed in reality. The design of such a simulation environment requires a compromise between the realism of biomechanical models used, the accuracy and stability of algorithms and solution methods implemented and the computational speed required for real-time haptic rendering. Modeling Mechanical restraint involves the use of an hyperelastic model (St Venant Kirchhoff) and a specific dynamic finite element code (mass tensor formulation). The different results are compared to experimental data. This comparison shows the effectiveness of the proposed physically based model.

International audience; Microinjection is the highly efficient delivery method of exogenousmaterials into cells, and it has been widely used in biomedical researchareas such as transgenics and genomics. However, this direct injectiontask is time consuming and laborious, resulting in low throughputand poor reproducibility. This paper describes a telerobotic sharedcontrol (TSC) framework for the microinjection with high manipulationefficiencies, in which a micromanipulator is controlled by the sharedmotion commands of both the human operator (direct manipulation)and the autonomous controller. To determine the optimal gains betweenthe operator and the controller, we proposed a quantitative evaluationmethod using Fitts' and steering laws. The results showed that a40%−60% weighting on the human operator produced better performancefor both speed and accuracy of task completion, and suggested thatsome level of automation or human involvement is important for microinjectiontasks.

The Moderate Resolution Imaging Spectrometer (MODIS) monitors the earth in 36 spectral bands using a cross-track double-sided continuously rotating scan mirror. The imperfect calibration of the linear arrays of detectors and additional random noise in the internal calibration system induce detector-to-detector stripes, mirror side stripes and noisy stripes visible in most emissive bands. This artefact affects seriously the visual quality and radiometric integrity of measured data. Several approaches including fourier filtering [1,2], wavelet analysis [3,4] and statistical techniques such as moment matching or histogram matching have been used to reduce striping on MODIS Data [5,6,7]. Despite an extensive and diverse destriping litterature, most techniques display residual stripes if not strong distortion from the original image. In this paper, we introduce a robust destriping methodology based on a variational approach.

In current AFM-based nanomanipulation systems, the commercial position closed-loop controller for piezoelectric nanopositioning stages are implemented with success in a wide range of industrial applications. Even if these controllers operate with satisfactory nominal tracking performance, considerable attention has been focused on appropriate control strategies to compensate hysteresis, nonlinearities, drift and creep for high bandwidths and large scanning regimes. As these closed-loop controllers are very cost-effective, a special interest in robust plug-in compensators seems to be a solution. We proposed in this paper a robust plug-in compensator using the H-infinity loop-shaping techniques which can be plugged into the existing controller without affecting the already satisfactory nominal tracking performance of the existing closed-loop system. Dynamic modeling, identification and robust control of a 3 d.o.f. piezoelectric nanorobotic positioner are presented in this paper in order to improve the nanorobot performance under plant parameter variations and in the presence of external disturbances. Simulation and experimental results are given to validate the proposed plug-in robust compensator in the case of a nanorobotic manipulation task.

International audience; To facilitate training of biological cell injection operations, we are developing an interactive virtual environment to simulate needle insertion into biological cells. A key challenge of deformable simulation is to satisfy the conflicting requirements of real-time interactivity and physical realism. This paper presents methodologies for dynamic modeling, visual/haptic display and model validation of cell injection. We first investigate the challenging issues in the modeling of the bio-mechanical properties of living cells. We propose a dynamic model to simulate cell deformation and puncture. The developed approach is based on the assumptions that the mechanical response of living cells is mainly determined by the cytoskeleton and that the cytoskeleton is organised as a tensegrity structure including microfilaments, microtubules and intermediate filaments. Equivalent microtubules struts are represented with a linear mass-tensor finite element model and equivalent microfilaments and intermediate filaments with viscoelastic Kelvin-Voigt elements. The virtual environment has been implemented with both graphic and haptic interfaces.

The control and monitoring of drinking water is becoming more and more interesting because of its effects on human life. Many techniques were developed in this field in order to ameliorate this process control attending to rigorous follow-ups of the quality of this vital resource. Several methods were implemented to achieve this goal. In this paper, a comparative study of two techniques resulting from the field of the artificial intelligence namely: RBF neural network (RBF-NN) and support vector machine (SVM), is presented. Developed from the statistical learning theory, these methods display optimal training performances and generalization in many fields of application, among others the field of pattern recognition. Applied as classification tools, these techniques should ensure within a multi-sensor monitoring system, a direct and quasi permanent control of water quality. In order to evaluate their performances, a simulation using real data, corresponding to the recognition rate, the training time, and the robustness, is carried out. To validate their functionalities, an application is presented.

We address the problem of computing a local orientation map in a digital image. We show that standard image gray level quantization causes a strong bias in the repartition of orientations, hindering any accurate geometric analysis of the image. In continuation, a simple dequantization algorithm is proposed, which maintains all of the image information and transforms the quantization noise in a nearby Gaussian white noise (we actually prove that only Gaussian noise can maintain isotropy of orientations). Mathematical arguments are used to show that this results in the restoration of a high quality image isotropy. In contrast with other classical methods, it turns out that this property can be obtained without smoothing the image or increasing the signal-to-noise ratio (SNR). As an application, it is shown in the experimental section that, thanks to this dequantization of orientations, such geometric algorithms as the detection of nonlocal alignments can be performed efficiently. We also point out similar improvements of orientation quality when our dequantization method is applied to aliased images.

The modern techniques in control and monitoring of drinking water, acquires a particular attention in the last few years. We attend more and more rigorous follow-ups of the quality of this resource, in order to master an effective control of the risks incurred for the public health. Several methods of control were implemented to meet this aim. In this paper, we present a comparative study of two techniques resulting from the field of the artificial intelligence namely: artificial neural networks (ANN), and support vector machines (SVM). Developed from the statistical learning theory, these methods display optimal training performances and generalization in many fields of application, among others the field of pattern recognition. Applied as classification tools, these techniques should ensure within a multi-sensor monitoring system, a direct and quasi permanent control of water quality. In order to evaluate their performances, a simulation corresponding to the recognition rate, the training time, and the robustness, is carried out. To validate their functionalities, an application of control of drinking water quality is presented.

International audience; We study the problem of finding the characteristic scale of a given satellite image. This feature is defined so that it does not depend on the spatial resolution of the image. This is a different problem than achieving scale invariance, as often studied in the literature. Our approach is based on the use of a linear scale-space and the total variation. The critical scale is defined as the one at which the normalized total variation reaches its maximum. It is shown experimentally, both on synthetic and real data, that the computed characteristic scale is resolution independent.

The field of monitoring drinking water acquires a particular importance in the last few years. The control of risks in the factories that produce and distribute water ensures the quality of this vital resource. Several methods and techniques were implemented in order to reduce these risks. We present here by a new technique called: support vector machines (SVMs). This method is developed from the statistical learning theory, which displays optimal training performances and generalization in several fields, among others the field of pattern recognition. The exposed technique ensures within a monitoring system, a direct and quasi permanent quality control of water. For a validation of the performances of this technique used as classification tool, a study in simulation of the training time, the recognition rate and the noise sensitivity, is carried out. With an aim of showing its functionality, an application test is presented.

This paper presents a new three-dimensional biomicromanipulation system for biological objects such as embryos, cells or oocytes. As the cell is very small, kept in the liquid, and observed through a microscope, the two-dimensional visual feedback makes difficult accurate manipulation in the 3-D world. To improve the manipulation work, we proposed an augmented human-machine interface. A 3-D visual information is provided to the operator through a 3-D reconstruction method using vision-based tracking deformations of the cell embryo. In order to stable injection tasks, the operator needs force feedback and haptic assistance during penetrating the cell envelop, the chorion. The proposed human-machine user's interface allows real-time realistic visual and haptic control strategies for constrained motion in image coordinates. Virtual haptic rendering allows to constrain the path of insertion and removal in the 3-D scene or to avoid cell destruction by controlling adequately position, velocity and force parameters

We study the problem of finding the characteristic scale of a given satellite image. We want to define this feature so that it does not depend on the spatial resolution of the image. Our approach is based on the use of a linear scale space and the total variation. The critical scale is defined as the one at which the normalized total variation is maximum.