A5062777204- Sparse and Compressive Sensing Techniques
- Image and Signal Denoising Methods
- Photoacoustic and Ultrasonic Imaging
- Numerical methods in inverse problems
- Advanced MRI Techniques and Applications
- Medical Imaging Techniques and Applications
- Advanced Image Processing Techniques
- Statistical Methods and Inference
- Blind Source Separation Techniques
- Advanced Optimization Algorithms Research
- Radiomics and Machine Learning in Medical Imaging
- Advanced Fluorescence Microscopy Techniques
- Markov Chains and Monte Carlo Methods
- Stochastic Gradient Optimization Techniques
- Gaussian Processes and Bayesian Inference
- Computational Drug Discovery Methods
- Remote-Sensing Image Classification
- Medical Image Segmentation Techniques
- Neural Networks and Applications
- Bayesian Modeling and Causal Inference
- Cell Image Analysis Techniques
- Spectroscopy and Chemometric Analyses
- Domain Adaptation and Few-Shot Learning
- Control Systems and Identification
- Face and Expression Recognition
Université Paris-Saclay
2019-2025
Inria Saclay - Île de France
2019-2025
Université Paris Cité
2011-2024
CentraleSupélec
2017-2024
Bouygues (France)
2017-2024
École Polytechnique
2018-2024
Institut national de recherche en informatique et en automatique
2018-2024
Laboratoire de Recherche en Informatique
2022
Center for Visual Communication (United States)
2022
Centre National de la Recherche Scientifique
2011-2021
The SARS-COV-2 pandemic has put pressure on intensive care units, so that identifying predictors of disease severity is a priority. We collect 58 clinical and biological variables, chest CT scan data, from 1003 coronavirus-infected patients two French hospitals. train deep learning model based scans to predict severity. then construct the multimodal AI-severity score includes 5 variables (age, sex, oxygenation, urea, platelet) in addition model. show neural network analysis CT-scans brings...
Modern signal processing (SP) methods rely very heavily on probability and statistics to solve challenging SP problems. are now expected deal with ever more complex models, requiring sophisticated computational inference techniques. This has driven the development of statistical based stochastic simulation optimization. Stochastic optimization algorithms computationally intensive tools for performing in models that analytically intractable beyond scope deterministic methods. They have been...
The l1/l2 ratio regularization function has shown good performance for retrieving sparse signals in a number of recent works, the context blind deconvolution. Indeed, it benefits from scale invariance property much desirable context. However, raises some difficulties when solving nonconvex and nonsmooth minimization problems resulting use such penalty term current restoration methods. In this paper, we propose new based on smooth approximation to function. addition, develop proximal-based...
In this work, we consider a class of differentiable criteria for sparse image computing problems, where nonconvex regularization is applied to an arbitrary linear transform the target image. As special cases, it includes edge-preserving measures or frame-analysis potentials commonly used in processing. shown by our asymptotic results, $\ell_2-\ell_0$ penalties may be employed provide approximate solutions $\ell_0$-penalized optimization problems. One advantages proposed approach that allows...
Variational methods are widely applied to ill-posed inverse problems for they have the ability embed prior knowledge about solution.However, level of performance these significantly depends on a set parameters, which can be estimated through computationally expensive and timeconsuming methods.In contrast, deep learning offers very generic efficient architectures, at expense explainability, since it is often used as black-box, without any fine control over its output.Deep unfolding provides...
The Poisson--Gaussian model can accurately describe the noise present in a number of imaging systems. However most existing restoration methods rely on approximations statistics. We propose convex optimization strategy for reconstruction images degraded by linear operator and corrupted with mixed noise. originality our approach consists considering exact, continuous-discrete corresponding to data After establishing Lipschitz differentiability convexity neg-log-likelihood, we derive...
In this work we address the problem of short-term load forecasting. We propose a generalization linear state-space model where evolution state and observation matrices is unknown. The proposed blind Kalman filter algorithm proceeds via alternating estimation these unknown inference state, within framework expectation-maximization. A mini-batch processing strategy introduced to allow on-the-fly experimental results show that method outperforms state-of-the-art techniques by considerable...
In brain MRI radiomics studies, the non-biological variations introduced by different image acquisition settings, namely scanner effects, affect reliability and reproducibility of results. This paper assesses how preprocessing methods (including N4 bias field correction resampling) harmonization (either six intensity normalization working on images or ComBat method radiomic features) help to remove effects improve feature in radiomics. The analyses were based vitro datasets (homogeneous...
Gliomas are among the most common types of central nervous system (CNS) tumors. A prompt diagnosis glioma subtype is crucial to estimate prognosis and personalize treatment strategy. The objective this study was develop a radiomics pipeline based on clinical Magnetic Resonance Imaging (MRI) scans noninvasively predict subtype, as defined tumor grade, isocitrate dehydrogenase (IDH) mutation status, 1p/19q codeletion status. total 212 patients from public retrospective Cancer Genome Atlas Low...
Co-administration of two or more drugs simultaneously can result in adverse drug reactions. Identifying drug-drug interactions (DDIs) is necessary, especially for development and repurposing old drugs. DDI prediction be viewed as a matrix completion task, which factorization (MF) appears suitable solution. This paper presents novel Graph Regularized Probabilistic Matrix Factorization (GRPMF) method, incorporates expert knowledge through graph-based regularization strategy within an MF...
This paper deals with the reconstruction of T1-T2 correlation spectra in nuclear magnetic resonance relaxometry. The ill-posed character and large size this inverse problem are main difficulties to tackle. While maximum entropy is retained as an adequate regularization approach, choice efficient optimization algorithm remains a challenging task. Our proposal apply truncated Newton two original features. First, theoretically sound line search strategy suitable for function applied ensure...
This monocentric retrospective study leveraged 200 multiparametric brain MRIs acquired between November 2019 and February 2020 at Gustave Roussy Cancer Campus (Villejuif, France). A total of 145 patients were included: 107 formed the training sample (55 ± 14 years, 58 women) 38 separate test (62 12 22 women). Patients had glioma, metastases, meningioma, or no enhancing lesion. T1, T2-FLAIR, diffusion-weighted imaging, low-dose, standard-dose postcontrast T1 sequences acquired. deep network...
Background The development and clinical adoption of quantitative imaging biomarkers (radiomics) has established the need for identification parameters altering radiomics reproducibility. aim this study was to assess impact magnetic field strength on resonance (MRI) features in neuroradiology practice. Methods T1 3D SPGR sequence acquired two phantoms 10 healthy volunteers with MR devices from same manufacturer using different fields (1.5 3T). Phantoms varied terms gadolinium concentrations...
This paper proposes accelerated subspace optimization methods in the context of image restoration. Subspace belong to class iterative descent algorithms for unconstrained optimization. At each iteration such methods, a stepsize vector allowing best combination several search directions is computed through multidimensional search. It usually obtained by an inner second-order method ruled stopping criterion that guarantees convergence outer algorithm. As alternative, we propose original...
Hyperspectral data unmixing aims at identifying the components (endmembers) of an observed surface and determining their fractional abundances inside each pixel area. Assuming that spectral signatures have been previously determined by endmember extraction algorithm, or to be part available library, main problem is reduced estimation abundances. For large hyperspectral image sets, abundance maps requires resolution a large-scale optimization subject linear constraints such as non-negativity...
Stochastic approximation techniques play an important role in solving many problems encountered machine learning or adaptive signal processing. In these contexts, the statistics of data are often unknown a priori their direct computation is too intensive, and they have thus to be estimated online from observed signals. For batch optimization objective function being sum fidelity term penalization (e.g., sparsity promoting function), Majorize-Minimize (MM) methods recently attracted much...
In this paper, a methodology is investigated for signal recovery in the presence of non-Gaussian noise. contrast with regularized minimization approaches often adopted literature, our algorithm regularization parameter reliably estimated from observations. As posterior density unknown parameters analytically intractable, estimation problem derived variational Bayesian framework where goal to provide good approximation distribution order compute mean estimates. Moreover, majorization...