Convex Minimization ; Sparsity priors; Variational Segmentation; Fetal imaging; Super-resolution Reconstruction; Magnetic Resonance Imaging; Inverse problem
Tourbier Sébastien, Velasco-Annis Clemente, Taimouri Vahid, Hagmann Patric, Meuli Reto, Warfield Simon K., Bach Cuadra Meritxell, Gholipour Ali (2017), Automated template-based brain localization and extraction for fetal brain MRI reconstruction, in NeuroImage
, 155, 460-472.
Tourbuer S., Schaer M., Warfield S.K., Meuli R., Gholipour A., Bach Cuadra M. (2016), Quantification of Fetal Cortical Folding using Slice-to-Volume Reconstructed MRI and FreeSurfer, in Proceeding of OHBM 22nd Annual Meeting
, Organization for Human Brain Mapping, Geneva.
Tourbier Sébastien, Bresson Xavier, Hagmann Patric, Thiran Jean Philippe, Meuli Reto, Bach Cuadra Meritxell (2015), An efficient total variation algorithm for super-resolution in fetal brain MRI with adaptive regularization, in NeuroImage
Tourbier Sébastien, Hagmann Patric, Cagneaux Maud, Guibaud Laurent, Gorthi Subrahmanyam, Schaer Marie, Thiran Jean Philippe, Meuli Reto, Bach Cuadra Meritxell (2015), Automatic brain extraction in fetal MRI using multi-atlas-based segmentation, in Proc. SPIE 9413, Medical Imaging 2015: Image Processing
, Orlando, Florida, United StatesProc. SPIE 9413, Medical Imaging 2015: Image Processing, Orlando, Florida, United States.
Tourbier Sébastien, Taimouri Vahid, Hagmann Patric, Velasco-Annis Clemente, Meuli Reto, Warfield Simon, Bach Cuadra Meritxell, Gholipour Ali (2015), Fully Automated Fetal Brain MRI Reconstruction, in INTELLIGENT IMAGING LINKING MR ACQUISITION AND PROCESSING Workshop
, Munich, GermanyMICCAI, Munich, Germany.
Tourbier Sébastien, Bresson Xavier, Hagmann Patric, Cagneaux Maud, Schaer Marie, Guibaud Laurent, Thiran Jean-Philippe, Meuli Reto, Bach Cuadra Meritxell (2014), Automated brain extraction in fetal MRI by multi-atlas fusion strategy: Study on healthy and pathological subjects, in 22nd of the International Society for Magnetic Resonance in Medicine
, Milano, ItalyProceedings of the International Society for Magnetic Resonance in Medicine, Milano, Italy.
Tourbier Sébastien, Bresson Xavier, Hagmann Patric, Thiran Jean Philippe, Meuli Reto A., Bach Cuadra Meritxell (2014), Efficient total variation algorithm for fetal brain MRI reconstruction, in Medical Image Computing and Computer-Assisted Intervention – MICCAI 2014
, Boston, USASpringer Lecture Notes in Computer Science (LNCS), Boston, USA.
Gianoni Martina, Schaer Marie, Tourbier Sébastien, Vial Yvan, Cagneaux Maud, Hagmann Patric, Meuli Reto, Bach Cuadra Meritxell (2014), In-vivo 3D Magnetic Resonance Volumetric Analysis of Fetal Cerebellum: From normal to pathology (unilateral cerebellar hypoplasia), in 22nd International Society for Magnetic Resonance in Medicine
, MilanProceedings of the International Society for Magnetic Resonance in Medicine, Milan, Italy.
Cagneaux Maud, Bach Cuadra Meritxell, Tourbier Sébastien, Schaer Marie, Hannoun S., Guibaud Laurent, Sappey-Marinier D. (2014), Segmentation of fetal pericerebral spaces based on reconstructed high-resolution MRI, in 22nd of the International Society for Magnetic Resonance in Medicine
, Milan, ItalyProceedings of the International Society for Magnetic Resonance in Medicine, Italy.
Ciurte Anca, Bresson Xavier, Cuisenaire Olivier, Houhou Nawal, Nedevschi Sergiu, Thiran Jean-Philippe, Bach Cuadra Meritxell (2014), Semi-supervised segmentation of ultrasound images based on patch representation and continuous min cut., in PloS one
, 9(7), 100972-100972.
Bresson Xavier, Laurent Thomas, Uminsky David, Von Brecht James H. (2013), Multiclass total variation clustering, in Advances in Neural Information Processing Systems 26
, USABurges, C., USA.
The last decades witnessed an impressive development of image rocessing techniques devoted to the study of the human brain. Along with technical improvements and increasing sample size of the studied population, the study of the adult brain anatomy and function progressively shifted to the assessment of earliest stages of cerebral development (i.e. the postnatal brain growth). If most of neuroimaging research studies are conducted in children and adults, recent advances in clinical imaging of the fetus provide an unprecedented opportunity to image the process of the human brain growth in utero. Quantitative studies of structural fetal Magnetic Resonance Imaging (MRI) would be of major importance in addressing the fundamental neuroscience question of early brain development. Moreover, in a clinical perspective, quantitative analysis of fetal MRI would help to better characterize the timing and the nature of prenatal pathologies, like ventriculomegaly, intra-uterine growth restriction or brain alterations in fetuses with congenital heart disease. New fast multi-slice MR techniques allow high contrast imaging of brain tissues but clinical acquisitions still have many critical limitations that restrain the use of computer-based methods for large scale studies. Consequently, the research project presented here, oriented to robust and accurate image processing methods for fetal MRI reconstruction and segmentation, will have an important impact for both neuroscience and clinics. Image processing methods for fetal MRI have to face major challenges like motion artifacts due to fetal movements inside the amniotic cavity, a poor spatial resolution due to fast sequences, the partial volume effect, intensity inhomogeneities, natural local intensity variations and the rapid changing anatomy of the developing fetal brain. The stateof-the-art of reconstruction and segmentation dedicated to fetal MRI is scarce. Reconstruction methods mostly rely on registration-interpolation strategies. However, these methods lack of mathematical proof of convergence and their resultspresent excessive blurring due to scattered data interpolation techniques. State-of-the-art segmentation methods are mostly applied to young fetus and they need anatomical priors and dynamic atlases to succeed. The use of such strong prior information faces a risk of circularity: each brain will be analyzed and deformed using the template of its biological age, potentially biasing the effective developmental delay. Thus, there is a need to develop solid mathematical framework for the fetal MRI reconstruction as well as for robust and accurate segmentation methods of fetuses at second and third trimester of pregnancy.The primary aim of this project resides in developing advanced fetal MR imaging processing, providing significantly improved reconstruction and segmentation methods as compared to state-of-the-art techniques. The originality of our research project resides in the use of soft priors in both reconstruction and segmentation problems. By soft priors, we mean prior information coming from the image data itself, like sparsity constrains, local spatial priors or label priors. Our first contribution will be a general and flexible reconstruction framework for fetal MRI, thanks to the large versatility of the sparsity constraints and of the convex optimization methods. We aim at solving the inverse problem formulation with regularization constraints like Total Variation (TV) or Generalized TV regularization. We are supported by results in other computer vision domains, like denoising or Super Resolution (SR), where these sparsity priors proven sharp edges reconstructed images while being robust to noise and motion-corrupted frames. The problem we are facing in fetal MRIis though more complex and thus further research will be done here, mainly dedicated to the development of iterative schemes and fast minimization algorithms to solve the 3D fetal MRI reconstruction. As regards segmentation, we aim at performing atlas-free segmentation of fetal brain tissue in MRI to avoid the risk of circularity in further quantitativestudies. We will present a flexible non-local variational framework based on a graph representation of the image with patches were any kind of regularization priors can be included, from new sparsity constraints like non-local TV to data driven priors. The use of patches (non-local information) is particularly appropriated in fetal MRI where we aim at preserving tiny structures and well-adapted to any gestational age. The proposed novel segmentation method will represent the second major methodological contribution of this project.