WP-4 Multidimensional Image Processing

Given the data produced in WP1 and WP5 or required by WP5 models and simulations, WP4 has two main objectives: to improve the imaging process in emerging imaging technics and to extract relevant imaging indicators and biomarkers in order to improve diagnosis and therapy in targeted medical applications.

Data processing is an important step for image reconstruction quality from various image production systems. The developments in WP4 aim at improving image formation in a number of imaging modalities where the measurement deficit may be significant (parsimonious data). These include developing new image reconstruction techniques (compressed sampling, reconstruction under constraint, stochastic optimization), proposing new inversion formulas and new reconstruction schemes for emerging imaging (X-ray CT, Spectral Computed Tomography, Fluorescence Optical Tomography, Compton Imaging for Hadron Therapy) and to develop generic methods to treat various types of noise.

The objectives of diagnosis and therapy lead to the development of methods for extracting image parameters and to propose new indicators and biomarkers for targeted pathologies. Robust and (semi-) automatic detection of structures or elementary events in images remains an open problem, given the multiple levels of uncertainty in imaging data (sparse and noisy data in particular). It is a question of studying new representations adapted to scalar, vectorial, tensorial, quaternionic data (multi-resolution combinatorial discrete models for 3D images and meshes, operators for Clifford's algebra and multi-quaternionic, nD analytical signal , fractal and multi-fractal analysis), to extend multi-dimensional operators for segmentation (sets of statistical levels, atlas-based, graph-cuts), multi-parametric estimation, registration and motion estimation (based on phase, optimal and suboptimal transport). Several aspects of decision support systems (supervised / unsupervised approaches, heterogeneous parameters, validation strategies) are also part of our work.

The developments carried out in WP4 are applied more particularly in cardiovascular imaging, brain imaging, bone imaging and image-guided cancer treatment.
 

• Alina TOMA 2012-2015 - Joint super-resolution/segmentation approaches for the tomographic images analysis of the bone micro-architecture
• Denis BUJOREANU 2015-2018 - Echographie compressée : une nouvelle stratégie d'acquisition et de formation pour une imagerie ultra-rapide
• Mathilde GIACALONE 2014-2017 - Traitement et simulation d’images d’IRM de perfusion pour la prédiction de l’évolution de la lésion ischémique dans l’accident vasculaire cérébral
• Meriem El Azami 2013-2016 - Computer aided diagnosis of epilepsy lesions based on multivariate and multimodality data analysis
• Nathan PAINCHAUD 2020-2023 - Apprentissage profond de variétés pour une meilleure caractérisation de l’hypertension artérielle en imagerie échocardiographique
• Nolann LAINE 2021 - 2024 - Analyse structurelle et cinématique de la paroi artérielle, dans les séquences d’images échographiques, par apprentissage profond
• Sarah LECLERC 2016-2019 - Automatisation de la segmentation sémantique de structures cardiaques en imagerie ultrasonore par apprentissage supervisé
• Sébastien CROMBEZ 2019-2022 - Hyperspectrale compressive imaging by deep convolutional neural network
• Sophie CARNEIRO-ESTEVES 2020-2023 - Segmentation des vaisseaux sanguins par approche variationnelle et apprentissage profond
• Tom HOHWEILLER 2016-2019 - Méthodes de décomposition non-linéaire pour l'imagerie X spectrale
• Yunyun SUN 2018 - 2021 - Patient-based color Doppler echocardiographic simulation