Showing papers by "David Rousseau published in 2016"

Deep learning based root-soil segmentation from X-ray tomography

Abstract: One of the most challenging computer vision problem in plant sciences is the segmentation of root and soil from X-ray tomography. So far, this has been addressed from classical image analysis methods. In this paper, we address this root/soil segmentation problem from X-ray tomography using a new deep learning classification technique. The robustness of this technique, tested for the first time on this plant science problem, is established with root/soil presenting a very low contrast in X-ray tomography. We also demonstrate the possibility to segment efficiently root from soil while learning on purely synthetic soil and root.

A Sensitive and Automatic White Matter Fiber Tracts Model for Longitudinal Analysis of Diffusion Tensor Images in Multiple Sclerosis

Abstract: Diffusion tensor imaging (DTI) is a sensitive tool for the assessment of microstructural alterations in brain white matter (WM). We propose a new processing technique to detect, local and global longitudinal changes of diffusivity metrics, in homologous regions along WM fiber-bundles. To this end, a reliable and automatic processing pipeline was developed in three steps: 1) co-registration and diffusion metrics computation, 2) tractography, bundle extraction and processing, and 3) longitudinal fiber-bundle analysis. The last step was based on an original Gaussian mixture model providing a fine analysis of fiber-bundle cross-sections, and allowing a sensitive detection of longitudinal changes along fibers. This method was tested on simulated and clinical data. High levels of F-Measure were obtained on simulated data. Experiments on cortico-spinal tract and inferior fronto-occipital fasciculi of five patients with Multiple Sclerosis (MS) included in a weekly follow-up protocol highlighted the greater sensitivity of this fiber scale approach to detect small longitudinal alterations.

On the value of the Kullback---Leibler divergence for cost-effective spectral imaging of plants by optimal selection of wavebands

Abstract: The practical value of a criterion based on statistical information theory is demonstrated for the selection of optimal wavelength and bandwidth of low-cost lighting systems in plant imaging applications. Kullback---Leibler divergence is applied to the problem of spectral band reduction from hyperspectral imaging. The results are illustrated on various plant imaging problems and show similar results to the one obtained with state-of-the-art criteria. A specific interest of the proposed approach is to offer the possibility to integrate technological constraints in the optimization of the spectral bands selected.

Robust real time motion compensation for intraoperative video processing during neurosurgery

Abstract: A motion compensation method dedicated to intraoperative RGB video imaging in neurosurgery is presented in this work. The dedicated motion model proposed is based on subspace learning of the patient brain motion. The resolution method uses keypoints for a sparse, fast and robust estimation of the brain motion. Our results, obtained from in vivo data, show that our method is as accurate as standard motion estimation method while being much faster. It is also very robust to un-predicted events that can happen in the operative room and opens the way to intraoperative real time hemodynamics map during neurosurgery on human subjects.

Shape descriptors to characterize the shoot of entire plant from multiple side views of a motorized depth sensor

Abstract: A low-cost depth camera recently introduced is synchronized with a specially devised low-cost motorized turntable. This results in a low-cost motorized depth sensor, able to provide a large number of registered side views, which is exploited here for the quantitative characterization of the shoots of entire plants. A set of four new shape descriptors of the shoots, constructed from the depth images on multiple side views of the shoots of plants, is proposed. The four descriptors quantify effective volume, multiscale organization, spatial symmetries and lacunarity of the plants. The four descriptors are here defined, validated on synthetic scenes with known properties, and then applied on nine different-looking real plants to illustrate their abilities for quantitative characterization and comparison. The resulting motorized depth sensor and associated image processing open new perspectives to various plant science applications including plant growth and architecture monitoring, plant response to stresses or the assessment of aesthetic parameters for ornamental plants.

An unsupervised spatio-temporal regularization for perfusion MRI deconvolution in acute stroke

Abstract: We consider the ill-posed inverse problem encountered in perfusion magnetic resonance imaging (MRI) analysis due to the necessity of eliminating, via a deconvolution process, the imprint of the arterial input function on the MR signals. Until recently, this deconvolution process was realized independently voxel by voxel with a sole temporal regularization despite the knowledge that the ischemic lesion in acute stroke can reasonably be considered piecewise continuous. A new promising algorithm incorporating a spatial regularization to avoid spurious spatial artifacts and preserve the shape of the lesion was introduced [1]. So far, the optimization of the spatio-temporal regularization parameters of the deconvolution algorithm was supervised. In this communication, we evaluate the potential of the L-hypersurface method in selecting the spatio-temporal regularization parameters in an unsupervised way and discuss the possibility of automating this method. This is demonstrated quantitatively with an in silico approach using digital phantoms simulated with realistic lesion shapes.

Interest of non-negativity constraint in perfusion DSC-MRI deconvolution for acute stroke

Abstract: We develop a globally convergent deconvolution algorithm for perfusion dynamic susceptibility contrast MRI applied to stroke. This algorithm includes, in addition to temporal and spatial regularization terms, a non-negativity constraint. Experiments on real data show performance improvements with the non-negativity constraint in the temporal regularization context.

Etude de la résistance partielle de la carotte à Alternaria dauci par une approche de phénotypage multi critères

Abstract: La creation de varietes presentant un niveau de resistance fort et durable passe immanquablement par la mise en œuvre de methodes de phenotypage toujours plus discriminantes, reproductibles et pouvant etre mises en œuvre sur de grands effectifs. Les methodes actuelles de phenotypage font encore tres souvent appel a une notation visuelle des symptomes. Nous rapportons ici la mise en œuvre de differentes methodes de phenotypage alternatives permettant de caracteriser, soit la phase asymptomatique par l’utilisation de nouvelles technologies telles que la qPCR, la thermographie ou la mesure de fluorescence de chlorophylle, soit la phase symptomatique a l’aide de tests en conditions controlees. Les avantages et les limites de ces methodes sont abordes.

Imagerie conjointe RGBS couleur et speckle pour la caractérisation de semences végétales

Abstract: L’imagerie de speckle est une technique non invasive et non destructive utilisée pour la caractérisation de tissus biologiques. Dans le domaine du végétal, elle est exploitée notamment pour caractériser la croissance, la viabilité des plantes ou leur infestation par des pathogènes [1, 2]. Nous présentons ici un dispositif de vision par ordinateur couplant deux modalités d’imagerie : speckle plein-champ et couleur RGB, et visant une application originale qui est l’étude de l’imbibition dans les semences. L’imbibition des semences est un processus essentiel qui déclenche le passage d’une graine sèche inerte à une plante vivante, et dont la dynamique demeure encore largement méconnue. Le dispositif d’imagerie, illustré sur la Fig. 1, exploite pour les deux modalités une unique caméra 3×8 bits avec une résolution spatiale de 5 Mpixels suffisante pour la taille des échantillons biologiques imagés. Pour l’imagerie RGB, une lumière blanche incohérente est utilisée et peut fonctionner d’une part en rétro-éclairage donnant accès à un masque binaire des échantillons obtenu par un seuillage automatique, et d’autre part en mode spéculaire fournissant des mesures colorimétriques des échantillons. Pour le speckle, une illumination en lumière cohérente est assurée avec un laser vert à λ = 532 nm. Le dispositif, synchronisé et piloté par ordinateur, donne accès à différentes séquences d’illumination/acquisition possibles. Une séquence typique consiste en l’alternance d’une acquisition de deux images RGB en lumière incohérente blanche en mode rétro-éclairage puis spéculaire, suivie de l’acquisition d’une série de N images de speckle. L’activité de biospeckle perçoit les micromouvements à la surface des échantillons biologiques, que l’on peut relier aux flux d’eau entrants lors de l’imbibition des semences. Cette activité est quantifiée par le calcul de l’intercovariance normalisée IN entre l’image initiale de référence et les N − 1 images de la série. La Fig. 1(d) présente des évolutions typiques de l’activité de biospeckle, qui fournissent des informations pouvant servir à une caractérisation de l’imbibition, son démarrage, intensité, durée, selon les conditions et les espèces. On cherche aussi à explorer les corrélations possibles entre les micromouvements et les changements de teinte observés lors de l’imbibition. La Fig. 1(e) montre des changements de teinte significatifs qui surviennent à la surface de graines de betterave selon les stades durant toute la durée de l’imbibition.

Comparison of propagation-based phase contrast tomography and full-field optical coherence tomography on bone tissue

Abstract: The current huge development of new 3D microscopic techniques (synchrotron microtomography, optical coherence tomography, light sheet microscopy, …) opens a large variety of new perspectives for life sciences. The contrasts of these new microscopies are mostly well understood on samples of known material content such as those used in physics or instrumentation studies. The situation is different when it comes to the interpretation of the contrasts observed with complex heterogeneous media found in biology. Therefore determining which 3D microscopy technique is suited for which biological question is a topic of current interest (see [1,2] for instance in our group).In this communication, we propose a comparison of the contrast observed with full-field optical coherence tomography (OCT) and propagation-based phase contrast tomography (PCT) on bone tissue at similar spatial resolution. A first comparison of OCT with standard absorption microtomography was given in [3] for bones and we extend this comparison to PCT which is known to provide enhanced contrast on bones at multiple scales [4]. The contrast of both these techniques are a priori interesting to be compared since they both rely on discontinuities of refraction index. This produces phase shift in PCT which operates in the X-ray domain with a monochromatic beam (generated by a synchrotron) while this generates direct intensity reflexion with OCT which only resorts to white light in the visible domain.