Deliverables Statistical evaluation Pursuing the objectives for the statistical evaluation, a set of queries was developed and tested successfully in the GlauCAD database. Data merging was achieved between the queries results and the end-point visits of the available patients. Only the second objective of the WP (evaluation of the clinical validity of quantitative optic disc assessment, performed by computer assisted stereophotogrammetry system) was possible, considering the amount of data available. Two separate studies were conducted: - For each group of patients that achieved an end-point, a prospective analysis was performed based on the 3 examinations previous to and after the end-point visit in order to check for possible statistical significant changes to any of the parameters considered. The same statistical changes were checked for all the patients (Paired T-test or Wilcoxon Non-Paramentric test). Statistical differences between groups were also checked (T-test or Mann-Whitney test). - For all parameters showing statistical differences, graphical representation was performed throughout the visits and per group. Although no statistical results were achieved due to the small number of patients with data filled in the database, the statistical work plan exists and was approved for the GlauCAD consortium, establishing a platform to be used when data is made available. Program for 3D reconstruction of stereo fundus images The stereo reconstruction software contributed by FhG/IGD addresses the reconstruction of metrologically meaningful 3D data of surfaces viewed from two different positions with photographic cameras (colour images). The software covers the complete processing chain from uncalibrated digital stereo image pairs to depth maps associated with one of the input images. Two main techniques are used here of which the first is self-calibration of image pairs, thereby assuming either significant depth in the images or a telecentric camera model to be adequate. The second is a dense matching technique based on a block matching approach. The processing chain also involves several steps of pre-processing, which enable a fully automated processing of stereo image pairs. In addition, FhG/IGD developed algorithms for the alignment of two depth maps aiming at the comparison of two data sets acquired at different points in time. Due to the unknown absolute scale of the depth maps, the alignment also takes into The software is customised to the demands of the GlauCAD project, however, major independent modules of the software are used in different contexts, e.g. an active range finder device based on a stereo camera system. Vessel deformation detection software This result is a diagnosis support tool for the case of Glaucoma. It receives as input a pair of retinal images from the same patient, taken at two time instances of a few months apart and produces an output by classifying the patient in two classes: non-converter (normal) or converter (abnormal). The tool consists of two main modules. The first registers the two images, identifies the retinal vessels central lines and calculates the values of shape descriptors for user-selected vessels. The second module classifies the patient according to the change of the values of shape descriptors between the time intervals of the two images, into the two aforementioned classes. GLAUCAD - Glaucoma prevention by computer aided diagnostics In the framework of the GLAUCAD project the following software and algorithm were developed: - Automatic papilla detection software: This software contains a robust algorithm for detecting the papilla area within a colour fundus image automatically. Tests were performed on thousands of images with varying qualities. The results show a very good estimation of the papilla area for a very large fraction of the image database. - 3D alignment software (Matlab prototype): Due to non-constant exposure conditions of the stereo slides and parameter estimations within the stereo reconstruction process, the resulting topographies have different depth scaling. This fact complicates the objective to extract parameters, which are capable to indicate smallest changes over time within a sequence of topographies. As a post-calibration approach failed, the only chance was a registration (affine transformation) by using constant features. For this, the retina area outside the disc was used. The method is to use the first topography of a sequence as reference and to align the following topographies to this one. For the registration itself a grid around the papilla (disc) area is defined in the reference topography and the corresponding points in the topography to be registered are found by calculating correlation coefficients in the color images. To find the complete 3D co-ordinates the corresponding z values of the topography are used. The distance function to be minimized is the sum of squared distances between the corresponding 3D co-ordinates. The distances itself are functions of the transformation parameters (translation, rotation, scaling). - Program for 3D sphere approximation: A universal algorithm to approximate an implicitly defined surface to a cloud of 3D co-ordinates was developed. Given a (initial) set of parameters to be optimised, this algorithm calculates the closest distance of the points to the surface (normal direction). The squares of these distances are summed to a distance function, which is dependent on the parameters to be optimised. These parameters are separated in shape, translation and rotation parameters. Searching for OpenAIRE data... There was an error trying to search data from OpenAIRE No results available
