Etienne Decencière

Exudate detection in color retinal images for mass screening of diabetic retinopathy

The automatic detection of exudates in color eye fundus images is an important task in applications such as diabetic retinopathy screening. The presented work has been undertaken in the framework of the TeleOphta project, whose main objective is to automatically detect normal exams in a tele-ophthalmology network, thus reducing the burden on the readers. A new clinical database, e-ophtha EX, containing precisely manually contoured exudates, is introduced. As opposed to previously available databases, e-ophtha EX is very heterogeneous. It contains images gathered within the OPHDIAT telemedicine network for diabetic retinopathy screening. Image definition, quality, as well as patients condition or the retinograph used for the acquisition, for example, are subject to important changes between different examinations. The proposed exudate detection method has been designed for this complex situation. We propose new preprocessing methods, which perform not only normalization and denoising tasks, but also detect reflections and artifacts in the image. A new candidates segmentation method, based on mathematical morphology, is proposed. These candidates are characterized using classical features, but also novel contextual features. Finally, a random forest algorithm is used to detect the exudates among the candidates. The method has been validated on the e-ophtha EX database, obtaining an AUC of 0.95. It has been also validated on other databases, obtaining an AUC between 0.93 and 0.95, outperforming state-of-the-art methods.

A multiple-instance learning framework for diabetic retinopathy screening

A novel multiple-instance learning framework, for automated image classification, is presented in this paper. Given reference images marked by clinicians as relevant or irrelevant, the image classifier is trained to detect patterns, of arbitrary size, that only appear in relevant images. After training, similar patterns are sought in new images in order to classify them as either relevant or irrelevant images. Therefore, no manual segmentations are required. As a consequence, large image datasets are available for training. The proposed framework was applied to diabetic retinopathy screening in 2-D retinal image datasets: Messidor (1200 images) and e-ophtha, a dataset of 25,702 examination records from the Ophdiat screening network (107,799 images). In this application, an image (or an examination record) is relevant if the patient should be referred to an ophthalmologist. Trained on one half of Messidor, the classifier achieved high performance on the other half of Messidor (A(z)=0.881) and on e-ophtha (A(z)=0.761). We observed, in a subset of 273 manually segmented images from e-ophtha, that all eight types of diabetic retinopathy lesions are detected.

Morphological decomposition of the surface topography of an internal combustion engine cylinder to characterize wear

A surface topography decomposition methodology is presented. It decomposes a surface into three elements: reference surface (waviness and form); superficial roughness (related to friction and wear); and valleys (related to lubricant circulation and reservoirs). It is applied to cylinder liners from an internal combustion V6 engine from in order to remove form and waviness components. The study of the resulting superficial roughness component has allowed a precise wear characterization.

/sup 18/F-FDG PET images segmentation using morphological watershed: a phantom study

Segmentation of 18F-FDG PET images could be helpful for delineation of tumor volume in radiotherapy and patient follow-up. The most commonly implemented method on clinical workstations is maximum intensity thresholding, which is inappropriate for heterogeneous uptakes. Our aim was to develop and evaluate a more sophisticated segmentation method, based on the morphological watershed. We developed a segmentation method taking into account PET images characteristics. We evaluated it first on phantom images, using an integrated PET/CT unit and taking CT images as reference images. To simulate tumors in a background activity, we used 6 homogeneous spheres of various volumes in a cylindrical phantom and 3 heterogeneous cylinders in an anthropomorphic phantom. The quality of segmentation was evaluated in terms of volume, shape and position. We compared the results with a maximum intensity threshold segmentation method fitting the volume, taken as reference segmentation. A quantitation analysis completed the phantom study. For both phantom acquisitions, the segmentation obtained with the watershed based algorithm gave satisfying results with the index integrating volume, shape and position. Results considering this index were not significantly different from the reference segmentation (p > 0.5). Errors of volume recovery reached 18% for watershed segmentation. The quantitation analysis on phantoms highlighted partial volume effect, with an error of activity concentration measurement on segmented images ranging between 42% and 51%. Performances of the watershed method evaluated in this study were comparable with an optimized segmentation on phantom images. The quantitation recovery of PET regions with this method was similar with to other segmentation methods.

Image Filtering Using Morphological Amoebas

This article presents the use of anisotropic dynamic structuring elements, or amoebas, in order to build content-aware noise reduction filters. The amoeba is the ball defined by a special geodesic distance computed for each pixel, and can be used as a kernel for many kinds of filters and morphological operators. 1. Introduction Noise is possibly the most annoying problem in the field of image processing. There are two ways to work around it: either design particularly robust algorithms that can work in noisy environments, or try to eliminate the noise in a first step while losing as little relevant information as possible and consequently use a normally robust algorithm. There are of course many algorithms that aim at reducing the amount of noise in images. In mathematical morphology filters can be, broadly-speaking, divided into two groups: 1 alternate sequential filters based on morphological openings and clos-ings, that are quite effective but also remove thin elements such as canals or peninsulas. Even worse, they can displace the contours and thus create additional problems in a segmentation application.

Automatic 3D segmentation of multiphoton images: a key step for the quantification of human skin.

Multiphoton microscopy has emerged in the past decade as a useful noninvasive imaging technique for in vivo human skin characterization. However, it has not been used until now in evaluation clinical trials, mainly because of the lack of specific image processing tools that would allow the investigator to extract pertinent quantitative three‐dimensional (3D) information from the different skin components.

Imaging and 3D morphological analysis of collagen fibrils

The recent booming of multiphoton imaging of collagen fibrils by means of second harmonic generation microscopy generates the need for the development and automation of quantitative methods for image analysis. Standard approaches sequentially analyse two‐dimensional (2D) slices to gain knowledge on the spatial arrangement and dimension of the fibrils, whereas the reconstructed three‐dimensional (3D) image yields better information about these characteristics. In this work, a 3D analysis method is proposed for second harmonic generation images of collagen fibrils, based on a recently developed 3D fibre quantification method. This analysis uses operators from mathematical morphology. The fibril structure is scanned with a directional distance transform. Inertia moments of the directional distances yield the main fibre orientation, corresponding to the main inertia axis. The collaboration of directional distances and fibre orientation delivers a geometrical estimate of the fibre radius. The results include local maps as well as global distribution of orientation and radius of the fibrils over the 3D image. They also bring a segmentation of the image into foreground and background, as well as a classification of the foreground pixels into the preferred orientations. This accurate determination of the spatial arrangement of the fibrils within a 3D data set will be most relevant in biomedical applications. It brings the possibility to monitor remodelling of collagen tissues upon a variety of injuries and to guide tissues engineering because biomimetic 3D organizations and density are requested for better integration of implants.

Parsimonious Path Openings and Closings

Path openings and closings are morphological tools used to preserve long, thin, and tortuous structures in gray level images. They explore all paths from a defined class, and filter them with a length criterion. However, most paths are redundant, making the process generally slow. Parsimonious path openings and closings are introduced in this paper to solve this problem. These operators only consider a subset of the paths considered by classical path openings, thus achieving a substantial speed-up, while obtaining similar results. In addition, a recently introduced 1D opening algorithm is applied along each selected path. Its complexity is linear with respect to the number of pixels, independent of the size of the opening. Furthermore, it is fast for any input data accuracy (integer or floating point) and works in stream. Parsimonious path openings are also extended to incomplete paths, i.e., paths containing gaps. Noise-corrupted paths can thus be processed with the same approach and complexity. These parsimonious operators achieve a several orders of magnitude speed-up. Examples are shown for incomplete path openings, where computing times are brought from minutes to tens of milliseconds, while obtaining similar results.

Parametric Optimization of Periodic Textured Surfaces for Friction Reduction in Combustion Engines

The aim of this paper is to analyze the tribological performance of different periodic textures of liner surfaces under hydrodynamic lubrication conditions, for an engine application. The impact of the different parameters of the periodic texture on its tribological performance is studied. First, a brief description of the numerical model is given. Then, the friction prediction tool is applied to the simulated periodic textures. Finally, optimal shapes and values of periodic textures are drawn out for the considered application.

Efficient Geodesic Attribute Thinnings Based on the Barycentric Diameter

An attribute opening is an idempotent, anti-extensive and increasing operator, which removes from an image connected components which do not fulfil a given criterion. When the increasingness property is dropped, we obtain a—more general—attribute thinning. In this paper, we propose efficient grey scale thinnings based on geodesic attributes.Given that the geodesic diameter is time consuming, we propose a new geodesic attribute, the barycentric diameter to speed up the computation time. Then, we give the theoretical error bound between these two attributes, and we note that in practice, the barycentric diameter gives very similar results in comparison with the geodesic diameter. Finally, we present the algorithm with further optimisations, to obtain a 60× speed up.We illustrate the use of these thinnings in automated non-destructive material inspection: the detection of cracks. We discuss the advantages of these operators over other methods such as path openings or the supremum of openings with segments.