The purpose of super-resolution approaches is to overcome the hardware limitations and the clinical requirements of imaging procedures by reconstructing high-resolution images from low-resolution acquisitions using post-processing methods. Super-resolution techniques could have strong impacts on structural magnetic resonance imaging when focusing on cortical surface or fine-scale structure analysis for instance. In this paper, we study deep three-dimensional convolutional neural networks for the super-resolution of brain magnetic resonance imaging data. First, our work delves into the relevance of several factors in the performance of the purely convolutional neural network-based techniques for the monomodal super-resolution: optimization methods, weight initialization, network depth, residual learning, filter size in convolution layers, number of the filters, training patch size and number of training subjects. Second, our study also highlights that one single network can efficiently handle multiple arbitrary scaling factors based on a multiscale training approach. Third, we further extend our super-resolution networks to the multimodal super-resolution using intermodality priors. Fourth, we investigate the impact of transfer learning skills onto super-resolution performance in terms of generalization among different datasets. Lastly, the learnt models are used to enhance real clinical low-resolution images. Results tend to demonstrate the potential of deep neural networks with respect to practical medical image applications.

International audienceRigid transformations are involved in a wide range of digital image processing applications. When applied on such discrete images, rigid transformations are however usually performed in their associated continuous space, then requiring a subsequent digitization of the result. In this article, we propose to study rigid transformations of digital images as a fully discrete process. In particular, we investigate a combinatorial structure modelling the whole space of digital rigid transformations on any subset of Z^2 of size N * N. We describe this combinatorial structure, which presents a space complexity O (N^9) and we propose an algorithm enabling to build it in linear time with respect to this space complexity. This algorithm, which handles real (i.e. non-rational) values related to the continuous transformations associated to the discrete ones, is however defined in a fully discrete form, leading to exact computation

The analysis of thin curvilinear objects in 3D images is a complex and challenging task. In this article, we introduce a new, non-linear operator, called RORPO (Ranking the Orientation Responses of Path Operators). Inspired by the multidirectional paradigm currently used in linear filtering for thin structure analysis, RORPO is built upon the notion of path operator from mathematical morphology. This operator, unlike most operators commonly used for 3D curvilinear structure analysis, is discrete, non-linear and non-local. From this new operator, two main curvilinear structure characteristics can be estimated: an intensity feature, that can be assimilated to a quantitative measure of curvilinearity; and a directional feature, providing a quantitative measure of the structure's orientation. We provide a full description of the structural and algorithmic details for computing these two features from RORPO, and we discuss computational issues. We experimentally assess RORPO by comparison with three of the most popular curvilinear structure analysis filters, namely Frangi Vesselness, Optimally Oriented Flux, and Hybrid Diffusion with Continuous Switch. In particular, we show that our method provides up to 8 percent more true positive and 50 percent less false positives than the next best method, on synthetic and real 3D images.

Magnetic resonance angiography (MRA) has become a common way to study cerebral vascular structures. Indeed, it enables to obtain information on flowing blood in a totally non-invasive and non-irradiant fashion. MRA exams are generally performed for three main applications: detection of vascular pathologies, neurosurgery planning, and vascular landmark detection for brain functional analysis. This large field of applications justifies the necessity to provide efficient vessel segmentation tools. Several methods have been proposed during the last fifteen years. However, the obtained results are still not fully satisfying. A solution to improve brain vessel segmentation from MRA data could consist in integrating high-level a priori knowledge in the segmentation process. A preliminary attempt to integrate such knowledge is proposed here. It is composed of two methods devoted to phase contrast MRA (PC MRA) data. The first method is a cerebral vascular atlas creation process, composed of three steps: knowledge extraction, registration, and data fusion. Knowledge extraction is performed using a vessel size determination algorithm based on skeletonization, while a topology preserving non-rigid registration method is used to fuse the information into the atlas. The second method is a segmentation process involving adaptive sets of gray-level hit-or-miss operators. It uses anatomical knowledge modeled by the cerebral vascular atlas to adapt the parameters of these operators (number, size, and orientation) to the searched vascular structures. These two methods have been tested by creating an atlas from a 18 MRA database, and by using it to segment 30 MRA images, comparing the results to those obtained from a region-growing segmentation method.

HAL is a multidisciplinary open access archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers. L'archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d'enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

Purpose:To propose an atlas-based method that uses both phase and magnitude images to integrate anatomical information in order to improve the segmentation of blood vessels in cerebral phase-contrast magnetic resonance angiography (PC-MRA). Material and Methods:An atlas of the whole head was developed to store the anatomical information. The atlas divides a magnitude image into several vascular areas, each of which has specific vessel properties. It can be applied to any magnitude image of an entire or nearly entire head by deformable matching, which helps to segment blood vessels from the associated phase image. The segmentation method used afterwards consists of a topology-preserving, region-growing algorithm that uses adaptive threshold values depending on the current region of the atlas. This algorithm builds the arterial and venous trees by iteratively adding voxels that are selected according to their grayscale value and the variation of values in their neighborhood. The topology preservation is guaranteed because only simple points are selected during the growing process. Results:The method was performed on 40 PC-MRA images of the brain. The results were validated using maximumintensity projection (MIP) and three-dimensional surface rendering visualization, and compared with results obtained with two non-atlas-based methods. Conclusion:The results show that the proposed method significantly improves the segmentation of cerebral vascular structures from PC-MRA. These experiments tend to prove that the use of vascular atlases is an effective way to optimize vessel segmentation of cerebral images. MAGNETIC RESONANCE ANGIOGRAPHY (MRA) is a noninvasive process (1) that provides three-dimensional images of vascular structures. The two main kinds of techniques that have been designed to visualize venous and arterial blood vessels are time-of-flight (TOF) MRA (2) and phase-contrast (PC) MRA (3). These techniques are frequently used to study the vascular structures of the brain. Indeed, obtaining precise information about brain vessels is fundamental for planning and performing neurosurgical procedures, as well as for detecting pathologies such as aneurysms and stenoses.Several methods devoted to vascular network segmentation from three-dimensional angiographic data have been published during the last 15 years. They can be divided into eight categories, corresponding to the main strategies used to carry out the segmentation, as follows: filtering, mathematical morphology, regiongrowing, vessel tracking, differential analysis, deformable models, statistical analysis, and artificial intelligence. It should be noted that the proposed classification is not the only one that can be used. Indeed, many other criteria can be chosen to classify the existing algorithms, including automation, the kind of data being processed, centerline or whole-vessel detection, size of the searched vessels, general methods, or methods devoted to a particular organ. The following text describes only a small part of the existing methods for ...

The hit-or-miss transform (HMT) is a fundamental operation on binary images, widely used since forty years. As it is not increasing, its extension to grey-level images is not straightforward, and very few authors have considered it. Moreover, despite its potential usefulness, very few applications of the grey-level HMT have been proposed until now. Part I of this paper, developed hereafter, is devoted to the description of a theory leading to a unification of the main definitions of the grey-level HMT, mainly proposed by Ronse and Soille, respectively (part II will deal with the applicative potential of the grey-level HMT, which will be illustrated by its use for vessel segmentation from 3D angiographic data). In this first part, we review the previous approaches to the grey-level HMT, especially the supremal one of Ronse, and the integral one of Soille; the latter was defined only for flat structuring elements, but it can be generalized to non-flat ones. We present a unified theory of the grey-level HMT, which is decomposed into two steps. First a fitting associates to each point the set of grey-levels for which the structuring elements can be fitted to the image; as in Soille's approach, this fitting step can be constrained. Next, a valuation associates a final grey-level value to each point; we propose three valuations: supremal (as in Ronse), integral (as in Soille) and binary.

a b s t r a c tThe Fuzzy C-Means (FCM) algorithm is a widely used and flexible approach to automated image segmentation, especially in the field of brain tissue segmentation from 3D MRI, where it addresses the problem of partial volume effects. In order to improve its robustness to classical image deterioration, namely noise and bias field artifacts, which arise in the MRI acquisition process, we propose to integrate into the FCM segmentation methodology concepts inspired by the non-local (NL) framework, initially defined and considered in the context of image restoration. The key algorithmic contributions of this article are the definition of an NL data term and an NL regularisation term to efficiently handle intensity inhomogeneities and noise in the data. The resulting new energy formulation is then built into an NL-FCM brain tissue segmentation algorithm. Experiments performed on both synthetic and real MRI data, leading to the classification of brain tissues into grey matter, white matter and cerebrospinal fluid, indicate a significant improvement in performance in the case of higher noise levels, when compared to a range of standard algorithms.