In this work, we focus on the estimation of sea surface current using Automated identification system (AIS) data streams in the Mediterranean sea. We propose to use deep learning techniques to solve the associated ill-posed inverse problem, for methodological purpose we compare two differ-ent approaches, the first one relies on a physical constrained unsupervised technique whereas the seconds exploit a super-vised framework and a dataset of in-situ observation from HF Radar. Performances are evaluated using ground-truth measurement provided by drifting buyos and HF Radar over area of the Sicily channel. We show that both AIS-derived product outperform satellite-altimetry derived ones in terms of reconstruction criterion. When comparing the two learning framework, the use of supervised learning algorithms leads to the best performances.

Online atlasing, i.e., incrementing an atlas with new images as they are acquired, is key when performing studies on very large, or still being gathered, databases. Regular approaches to atlasing however do not focus on this aspect and impose a complete reconstruction of the atlas when adding images. We propose instead a diffeomorphic online atlasing method that allows gradual updates to an atlas. In this iterative centroid approach, we integrate new subjects in the atlas in an iterative manner, gradually moving the centroid of the images towards its final position. This leads to a computationally cheap approach since it only necessitates one additional registration per new subject added. We validate our approach on several experiments with three main goals: 1- to evaluate atlas image quality of the obtained atlases with sharpness and overlap measures, 2- to assess the deviation in terms of transformations with respect to a conventional atlasing method and 3- to compare its computational time with regular approaches of the literature. We demonstrate that the transformations divergence with respect to a state-of-the-art atlas construction method is small and reaches a plateau, that the two construction methods have the same ability to map subject homologous regions onto a common space and produce images of equivalent quality. The computational time of our approach is also drastically reduced for regular updates. Finally, we also present a direct extension of our method to update spatio-temporal atlases, especially useful for developmental studies.

Space oceanography missions, especially altimeter missions, have considerably improved the observation of sea surface dynamics over the last decades. They can however hardly resolve spatial scales below ~ 100km. Meanwhile the AIS (Automatic Identification System) monitoring of the maritime traffic implicitly conveys information on the underlying sea surface currents as the trajectory of ships is affected by the current. Here, we show that an unsupervised variational learning scheme provides new means to elucidate how AIS data streams can be converted into sea surface currents. The proposed scheme relies on a learnable variational framework and relate to variational auto-encoder approach coupled with neural ODE (Ordinary Differential Equation) solving the targeted ill-posed inverse problem. Through numerical experiments on a real AIS dataset, we demonstrate how the proposed scheme could significantly improve the reconstruction of sea surface currents from AIS data compared with state-of-the-art methods, including altimetry-based ones.

Variational models are among the state-of-the-art formulations for the resolution of ill-posed inverse problems. Following recent advances in learning-based variational settings, we investigate the end-to-end learning of variational models, more precisely of the regularization term given some observation model, jointly to the associated solver, so that we can optimize the reconstruction performance. In the proposed end-to-end setting, both the variational cost and the gradient-based solver are stated as neural networks using automatic differentiation for the latter. We consider an application to inverse problems with incomplete datasets (image inpainting and multivariate time series interpolation). We experimentally illustrate that this framework can lead to a significant gain in terms of reconstruction performance, including w.r.t. the direct minimization of the variational formulation derived from the known generative model.

This work deals with data-driven and learning-based approaches to fill space-time sampling gaps in the observation of sea surface tracers such as Sea Surface Height (SSH), Sea Surface temperature (SST), Ocean Colour,... More precisely, we jointly address field super-resolution and the prediction of subgrid-scale features, which is novel to our knowledge. From a methodological point of view, we consider deep learning architectures with a view to learning geophysically-sound differential operators (i.e. trasnport operators). Based on an Observing System Simulation Experiment representative of SWOT fast sampling phase using NATL60 simulation data, we illustrate the relevance of the proposed methodological framework which reconstructs more than 90% of the variance of the high-resolution SSH anomaly field and above 90% of the subgrid-scale variance of this anomaly. We also illustrate significant gain w.r.t. other baseline neural network architectures and further discuss the relevance of the reported contribution for other tracer fields and case studies.

Deep Learning (DL), especially Convolutional Neural Networks (CNN), has become the state-of-the-art for a variety of pattern recognition issues. Technological developments have allowed the use of high-end General Purpose Graphic Processor Units (GPGPU) for accelerating numerical problem solving. They resort no only to lower computational time, but also allow considering much larger networks. Hence, nowadays computers are able to drive deeper, wider and more powerful models. State of the art CNNs have achieved human-like performance in several recognition tasks such as: handwritten character recognition, face recognition, scene labelling, object detection and image classification among others. Meanwhile, mobile devices have become powerful enough to handle the computations required for deploying CNNs models in near real-time. Here, we investigate the implementation of light-weight CNN schemes on mobile devices for domain-specific objection recognition tasks.

We here address the super-resolution of a high-resolution image involving missing data given that a low-resolution image of the same scene is available. This is a typical issue in the remote sensing of geophysical parameters from different spaceborne sensors. Such super-resolution application involves large downscaling factor (typically from 10 to 20) and the super-resolution model should account for both texture patterns and specific statistical features, especially the spectral and non-Gaussian features. In this context, we propose a novel non-local approach and formally states the solution as the joint minimization of several projection constraints. We illustrate the relevance of the proposed model on real ocean remote sensing data, namely sea surface temperature fields, as well on visual textures.

Understanding the geometrical organization of the white matter fibers is one of the current challenges in neuroimaging. White matter fiber clustering technique appears to a corner stone to solve this problem. In this paper, we propose a rapid and efficient unsupervised white matter fiber tracts clustering methodology based on a novel fiber tract similarity metric and an approximation of the k-means algorithm. In this approach, we first define a distance metric capable to quantify the intrinsic geometry of the fiber tracts. This metric is based on a combination of the symmetric Chamfer distance and mean local orientation measures between fiber tracts. Second, we perform the randomized feature selection algorithm proposed for the k-means problem to reduce the dimensionality of the distance data matrix generated from all the fiber tracts using the defined metric. The k-means algorithm is then performed on the reduced distance matrix to cluster the fiber tracts. Finally, we evaluate the method on the synthetic data and in vivo adult brain dataset.

This paper presents an approach to 3-D diffusion tensor image (DTI) reconstruction from multi-slice diffusion weighted (DW) magnetic resonance imaging acquisitions of the moving fetal brain. Motion scatters the slice measurements in the spatial and spherical diffusion domain with respect to the underlying anatomy. Previous image registration techniques have been described to estimate the between slice fetal head motion, allowing the reconstruction of 3D a diffusion estimate on a regular grid using interpolation. We propose Approach to Unified Diffusion Sensitive Slice Alignment and Reconstruction (AUDiSSAR) that explicitly formulates a process for diffusion direction sensitive DW-slice-to-DTI-volume alignment. This also incorporates image resolution modeling to iteratively deconvolve the effects of the imaging point spread function using the multiple views provided by thick slices acquired in different anatomical planes. The algorithm is implemented using a multi-resolution iterative scheme and multiple real and synthetic data are used to evaluate the performance of the technique. An accuracy experiment using synthetically created motion data of an adult head and an experiment using synthetic motion added to sedated fetal monkey dataset show a significant improvement in motion-trajectory estimation compared to current state-of-the-art approaches. The performance of the method is then evaluated on challenging but clinically typical in utero fetal scans of four different human cases, showing improved rendition of cortical anatomy and extraction of white matter tracts. While the experimental work focuses on DTI reconstruction (second-order tensor model), the proposed reconstruction framework can employ any 5-D diffusion volume model that can be represented by the spatial parameterizations of an orientation distribution function.

We propose in this work a patch-based segmentation method relying on a label propagation framework. Based on image intensity similarities between the input image and a learning dataset, an original strategy which does not require any non-rigid registration is presented. Following recent developments in non-local image denoising, the similarity between images is represented by a weighted graph computed from intensity-based distance between patches. Experiments on simulated and in-vivo MR images show that the proposed method is very successful in providing automated human brain labeling.

A common solution to clinical MR imaging in the presence of large anatomical motion is to use fast multislice 2D studies to reduce slice acquisition time and provide clinically usable slice data. Recently, techniques have been developed which retrospectively correct large scale 3D motion between individual slices allowing the formation of a geometrically correct 3D volume from the multiple slice stacks. One challenge, however, in the final reconstruction process is the possibility of varying intensity bias in the slice data, typically due to the motion of the anatomy relative to imaging coils. As a result, slices which cover the same region of anatomy at different times may exhibit different sensitivity. This bias field inconsistency can induce artifacts in the final 3D reconstruction that can impact both clinical interpretation of key tissue boundaries and the automated analysis of the data. Here we describe a framework to estimate and correct the bias field inconsistency in each slice collectively across all motion corrupted image slices. Experiments using synthetic and clinical data show that the proposed method reduces intensity variability in tissues and improves the distinction between key tissue types.

We propose in this work a patch-based image labeling method relying on a label propagation framework. Based on image intensity similarities between the input image and an anatomy textbook, an original strategy which does not require any nonrigid registration is presented. Following recent developments in nonlocal image denoising, the similarity between images is represented by a weighted graph computed from an intensity-based distance between patches. Experiments on simulated and in vivo magnetic resonance images show that the proposed method is very successful in providing automated human brain labeling.

In recent years, postprocessing of fast multislice magnetic resonance imaging (MRI) to correct fetal motion has provided the first true 3-D MR images of the developing human brain in utero. Early approaches have used reconstruction based algorithms, employing a two-step iterative process, where slices from the acquired data are realigned to an approximate 3-D reconstruction of the fetal brain, which is then refined further using the improved slice alignment. This two step slice-to-volume process, although powerful, is computationally expensive in needing a 3-D reconstruction, and is limited in its ability to recover subvoxel alignment. Here, we describe an alternative approach which we term slice intersection motion correction (SIMC), that seeks to directly co-align multiple slice stacks by considering the matching structure along all intersecting slice pairs in all orthogonally planned slices that are acquired in clinical imaging studies. A collective update scheme for all slices is then derived, to simultaneously drive slices into a consistent match along their lines of intersection. We then describe a 3-D reconstruction algorithm that, using the final motion corrected slice locations, suppresses through-plane partial volume effects to provide a single high isotropic resolution 3-D image. The method is tested on simulated data with known motions and is applied to retrospectively reconstruct 3-D images from a range of clinically acquired imaging studies. The quantitative evaluation of the registration accuracy for the simulated data sets demonstrated a significant improvement over previous approaches. An initial application of the technique to studying clinical pathology is included, where the proposed method recovered up to 15 mm of translation and 30 degrees of rotation for individual slices, and produced full 3-D reconstructions containing clinically useful additional information not visible in the original 2-D slices.

This paper examines a novel approach for temporal calibration of a three-dimensional (3-D) freehand ultrasound system. A localization system fixed on the probe gives the position and orientation of the probe. For quantitative use, calibration is needed to correctly localize a B-scan in four-dimensional (4-D) (3-D+t) space. Temporal latency estimation is defined in a general robust formulation using no specific probe motion constraints. Experiments were performed on synthetic and real data using a 3-D freehand ultrasound system. The achieved precision is lower than the image acquisition rate (40 ms). A validation study using a calibration phantom has been performed to evaluate the influence of incorrect latency estimation on the 3-D reconstruction procedure. We showed that for latency estimation errors less than 40 ms, the 3-D reconstruction errors are negligible for volume estimation.

In this paper, three different calibration methods for three-dimensional (3-D) freehand ultrasound (US) are evaluated. Calibration is the process of estimating the rigid transformation from US image coordinates to the coordinate system of the tracking sensor mounted onto the probe. Calibration accuracy has an important impact on quantitative studies. Geometrical precision can also be crucial in many interventions and surgery. The proposed evaluation framework relies on a single point phantom and a 3-D US phantom which mimics the US characteristics of human liver. Four quality measures are used: 3-D point localization criterion, distance and volume measurements, and shape based criterion. Results show that during the acquisition procedure, volumetric measurements and shapes of the reconstructed object depend on probe motion used, particularly fan motions for which errors are larger. It is also shown that accurate calibration is essential to obtain reliable quantitative information.

A new registration method for ultrasound volumes relying on on a statistical texture-based similarity measure is investigated. Texture information is given by spatial Gabor filters and represented by statistical kernel-based distributions. The registration similarity measure is then defined as a probabilistic distance, derived from Bhattacharyya coefficient, between two statistical distributions. Given this similarity measure, parametric ultrasound image registration is stated as a robust minimization issue. We also exploit frequency properties of spatial Gabor filters to propose a multiresolution approach to perform this minimization. We provide a preliminary evaluation of the new registration technique on clinical data.