%0 Journal Article %J MEDICAL IMAGE ANALYSIS %D 2010 %T Glaucoma Risk Index: Automated glaucoma detection from color fundus images %A Rudriger Bock %A Jörg Meier %A László Gábor Nyúl %A Joachim Hornegger %A Georg Michelson %X

Glaucoma as a neurodegeneration of the optic nerve is one of themost common causes of blindness. Because revitalization of the degenerated nerve fibers of the optic nerve is impossible early detection of the disease is essential. This can be supported by a robust and automated mass-screening. We propose a novel automated glaucoma detection system that operates on inexpensive to acquire and widely used digital color fundus images. After a glaucoma specific preprocessing, different generic feature types are compressed by an appearance-based dimension reduction technique. Subsequently, a probabilistic two-stage classification scheme combines these features types to extract the novel Glaucoma Risk Index (GRI) that shows a reasonable glaucoma detection performance. On a sample set of 575 fundus images a classification accuracy of 80% has been achieved in a 5-fold cross-validation setup. The GRI gains a competitive area under ROC (AUC) of 88% compared to the established topography- based glaucoma probability score of scanning laser tomography with AUC of 87%. The proposed color fundus image-based GRI achieves a competitive and reliable detection performance on a low-priced modality by the statistical analysis of entire images of the optic nerve head.

%B MEDICAL IMAGE ANALYSIS %V 14 %P 471 - 481 %8 2010 %@ 1361-8415 %G eng %N 3 %9 Journal article %! MED IMAGE ANAL %0 Journal Article %J INVESTIGATIVE OPHTHALMOLOGY & VISUAL SCIENCE %D 2009 %T Multimodal Automated Glaucoma Detection Combining the Glaucoma Probability Score and the Glaucoma Risk Index %A Rudriger Bock %A Jörg Meier %A László Gábor Nyúl %A Joachim Hornegger %A Georg Michelson %X

Purpose:Fundus camera and Heidelberg Retina Tomograph (HRT) arecommonly used for reliable glaucoma diagnosis. Quantitative glaucoma scores, however, do not utilize both image content simultaneously. We propose the combination of topography and fundus image based indices for automated glaucoma detection which outperforms their sole application of either. Methods:The probabilistic values of topography based Glaucoma Probability Score (GPS) and our fundus image based Glaucoma Risk Index (GRI) are assembled to a two-dimensional feature space. In contrast to established methods the subsequent application of a probabilistic nu-Support Vector Machine classifier (nu = 0.5, kernel: radial basis function) uses both the topographic and the textural information to determine a final glaucoma probability. Instances labeled with a final probability greater than 0.5 are considered glaucomatous.For the evaluations in a 10-fold cross- validation setup, we took a sample set (mean age: 55.4 ± 10.9 years) of papilla images of 149 glaucomatous patients (FDT test time 67.4 ± 35.6 s) and 246 normals from the Erlangen Glaucoma Registry. The gold standard diagnosis was given by a glaucoma specialist based on an elaborate ophthalmological examination with ophthalmoscopy, visual field, IOP, FDT, and HRT II. The GPS was calculated by HRT device while papilla centered color fundus images (Kowa non-myd, FOV 22°) were used to calculate the GRI. Results:The classification of the GRI resulted in an area under ROC curve (AUC) of 0.81 with an F-measure of 0.71 for glaucomatous cases and 0.83 for normals. The GPS achieved an AUC of 0.86 while the F-measure for glaucoma was 0.74 (F-measure for healthy was 0.84).The combination of both indices clearly increased the AUC by 4% up to 0.9 compared to the sole application of the GPS. The F-measure for glaucomatous images was improved up to 0.76 (F-measure for healthy images was 0.86). Conclusions:The proposed combination of the topography based GPS and the fundus image based GRI shows superior performance compared to either index alone.Both indices utilize complementary information about the glaucoma disease. Consequently, this multimodal combined application of both indices is promising to reach a more reliable automated glaucoma detection performance. The approach can be used in large screening applications where an automated tool is essential to support the experts in finding glaucomatous eyes.

%B INVESTIGATIVE OPHTHALMOLOGY & VISUAL SCIENCE %V 50 %P 324 %8 2009 %@ 0146-0404 %G eng %N 5 %9 Abstract %! INVEST OPHTH VIS SCI %0 Journal Article %J INVESTIGATIVE OPHTHALMOLOGY & VISUAL SCIENCE %D 2008 %T Automated Glaucoma Detection From Color Fundus Photographs %A Rudriger Bock %A Jörg Meier %A László Gábor Nyúl %A Joachim Hornegger %A Georg Michelson %X

Purpose:The presentation of a novel fully automated system thatseparates glaucomatous from healthy cases based on digital fundus images. Methods:A pre-processing step eliminates certain disease independent variations such as illumination inhomogeneities, papilla size differences and vessel structures from the input images. In order to characterize glaucomatous changes, generic feature types (pixel intensities, frequency coefficients, histogram parameters, Gabor textures, spline coefficients) are extracted. In contrast to existing approaches, each feature vector is compressed by Principal Component Analysis. The classification of the transformed features is done by a state- of-the-art nu-Support Vector Machine.For the elaborate experimental evaluation of the proposed system architecture we took a large set of papilla-centered color fundus images of 100 glaucoma patients (FDT test time 67.25 ± 33.4 s) and 100 normals (overall mean age 57.0 ± 10.0 years) from the Erlangen Glaucoma Registry (Kowa non-myd, FOV 22,5°). The gold standard was given by an experienced ophthalmologist based on a complete ophthalmological examination with ophthalmoscopy, visual field, IOP, FDT, and HRT II. Results:Classification of compressed raw pixel intensities gained a success rate of 83% with a specificity of 0.72 and a sensitivity of 0.94 to detect glaucomatous cases. A success rate of 86% was achieved by using spline coefficients with a specificity of 0.78 and a sensitivity of 0.94 to detect glaucoma. The combination of both features slightly increased specificity to 0.82 (sensitivity = 0.92). The kappa statistic of 0.74 states a robust classification scheme. Conclusions:The proposed algorithm achieves a robust and competitive glaucoma detection rate. It is comparable to known methods applied to topographic papilla images and does not depend on segmentation-based measurements. For the first time, automated glaucoma detection is performed on color fundus images. Thus, fundus photography is an appropriate modality for computer-assisted glaucoma screening.

%B INVESTIGATIVE OPHTHALMOLOGY & VISUAL SCIENCE %V 49 %P 1863 %8 2008 %@ 0146-0404 %G eng %N 5 %9 Journal article %! INVEST OPHTH VIS SCI %0 Journal Article %J INVESTIGATIVE OPHTHALMOLOGY & VISUAL SCIENCE %D 2008 %T The Erlanger Glaucoma Matrix - A Visualization Approach Towards Optimal Glaucomatous Optic Nerve Head Image Presentation %A Jörg Meier %A Rudriger Bock %A C Forman %A László Gábor Nyúl %A Joachim Hornegger %A Georg Michelson %X

Purpose:Presentation of a two-dimensional visualization approachfor intuitive and reliable glaucoma diagnosis and for setting a current observation into a relationship with pre-diagnosed data. Methods:We present a new matrix visualization technique for digital optic nerve head images. The matrix is filled with 300 pre-diagnosed reference images which show different papilla sizes and varying stages of glaucoma disease. In matrix rows the samples range from healthy ones to advanced glaucoma cases. In matrix columns the papillas are ordered by the size of the optic nerve head. The approach generalizes such that the samples can be ordered by additional criteria, too, e. g. subjects' age or anamnestic risk factors. Furthermore arbitrary image modalities and image numbers can be incorporated. Results:The glaucoma classification of a single image is difficult even for experts. Our proposed visualization provides an intuitive way for neighborhood comparisons of optic nerve head images. It allows to evaluate an image in the context of given pre-diagnosed reference samples. By the two-dimensional presentation one can study disease-dependent changes separate from other variations. Glaucoma progression can be observed separated from size variations. Thus, it supports diagnosis even in problematic cases such as macropapillas. The trustworthiness of physicians' diagnosis can be improved. Conclusions:Our approach gives insights on glaucomatous optic nerve appearance in relation to varying papilla sizes. The novel visualization of a single image within the context of other images is considered as an important tool for learning and training medical glaucoma detection. This approach visualizes computer calculated risk estimations by presenting the result within context of given gold-standard images. In contrast to pure classification systems our method does not come up with a hard decision but explains the relationship to similar pre- diagnosed cases.

%B INVESTIGATIVE OPHTHALMOLOGY & VISUAL SCIENCE %I Arvo %V 49 %P 1893 %8 May 2008 %@ 0146-0404 %G eng %N 5 %9 Journal article %! INVEST OPHTH VIS SCI %0 Book Section %B Analysis of Biomedical Signals and Images %D 2008 %T Novel Visualization Approach of an Automated Image Based Glaucoma Risk Index for Intuitive Diagnosis %A Jörg Meier %A Rudriger Bock %A László Gábor Nyúl %A Joachim Hornegger %A Georg Michelson %E Jiří Jan %E Jiří Konzuplik %E Ivo Provazník %X

Glaucoma is one of the most common causes for blindnessworldwide. Screening is adequate to detect glaucoma at an early stage. Although it is supported by computer assisted tools no further information from former clinical studies is incorporated. We devised a novel visualization tool that presents additional comparative image data for the diagnosis process. Automated computation of a glaucoma risk index on color fundus photographs is used to initially position an undiagnosed image in reference data. The index achieves a competitive glaucoma detection rate. The combination of the automated risk index and the new visualization technique is an important tool towards a faster and more reliable diagnosis of glaucoma.

%B Analysis of Biomedical Signals and Images %I Brno University of Technology %C Brno %P 205 - 209 %8 2008/// %G eng %0 Book Section %B 3rd Russian-Bavarian Conference on Bio-Medical Engineering, Proceedings %D 2007 %T Appearance-based Approach to Extract an Age-related Biomarker from Retinal Images %A Rudriger Bock %A Jörg Meier %A László Gábor Nyúl %A Simone Wärntges %A Georg Michelson %A Joachim Hornegger %E Joachim Hornegger %E Ernst W Mayr %E Sergey Schookin %E Hubertus Feußner %E Nassir Navab %E Yuri V. Gulyaev %E Kurt Höller %E Victor Ganzha %X

We present an appearance-based method that extracts a new age-related biomarker from retina images. The Principal Component Analysis is applied on intensity values of the illumination corrected green channel of fundus images. The algorithm does not use segmentation, is robust and shows a high range of reliability. It identified an age-related feature with a strong influence of the temporal parapapillary area and the optic nerve head. The feature correlates with chronological age of the participants and is significantly influenced by the appearance of cardiovascular risk factors such as smoking and hypertension, and thus it can be designated a biomarker. We extract and validate a medical parameter from retina images applying a purely data-driven approach without using any prior knowledge.

%B 3rd Russian-Bavarian Conference on Bio-Medical Engineering, Proceedings %I Friedrich-Alexander University Erlangen-Nuremberg %C Erlangen %V 1 %P 127 - 131 %8 2007 %G eng %9 Conference paper %0 Book Section %B Pattern Recognition %D 2007 %T Classifying Glaucoma with Image-based Features from Fundus Photographs %A Rudriger Bock %A Jörg Meier %A Georg Michelson %A László Gábor Nyúl %A Joachim Hornegger %E Fred A Hamprecht %E Christoph Schnorr %E Bernd Jähne %X

Glaucoma is one of the most common causes of blindness and it isbecoming even more important considering the ageing society. Because healing of died retinal nerve fibers is not possible early detection and prevention is essential. Robust, automated mass-screening will help to extend the symptom-free life of affected patients. We devised a novel, automated, appearance based glaucoma classification system that does not depend on segmentation based measurements. Our purely data-driven approach is applicable in large-scale screening examinations. It applies a standard pattern recognition pipeline with a 2-stage classification step. Several types of image-based features were analyzed and are combined to capture glaucomatous structures. Certain disease independent variations such as illumination inhomogeneities, size differences, and vessel structures are eliminated in the preprocessing phase. The “vessel-free” images and intermediate results of the methods are novel representations of the data for the physicians that may provide new insight into and help to better understand glaucoma. Our system achieves 86 % success rate on a data set containing a mixture of 200 real images of healthy and glaucomatous eyes. The performance of the system is comparable to human medical experts in detecting glaucomatous retina fundus images.

%B Pattern Recognition %S Lecture Notes in Computer Science %I Springer Verlag %C Heidelberg %P 355 - 364 %8 Sep 2007 %@ 978-3-540-74933-2 %G eng %9 Conference paper %! LNCS %R 10.1007/978-3-540-74936-3_36 %0 Book Section %B Computer Analysis of Images and Patterns %D 2007 %T Effects of Preprocessing Eye Fundus Images on Appearance Based Glaucoma Classification %A Jörg Meier %A Rudriger Bock %A Georg Michelson %A László Gábor Nyúl %A Joachim Hornegger %E Walter G Kropatsch %E Martin Kampel %E Allan Hanbury %X

Early detection of glaucoma is essential for preventing one ofthe most common causes of blindness. Our research is focused on a novel automated classification system based on image features from fundus photographs which does not depend on structure segmentation or prior expert knowledge. Our new data driven approach that needs no manual assistance achieves an accuracy of detecting glaucomatous retina fundus images compareable to human experts. In this paper, we study image preprocessing methods to provide better input for more reliable automated glaucoma detection. We reduce disease independent variations without removing information that discriminates between images of healthy and glaucomatous eyes. In particular, nonuniform illumination is corrected, blood vessels are inpainted and the region of interest is normalized before feature extraction and subsequent classification. The effect of these steps was evaluated using principal component analysis for dimension reduction and support vector machine as classifier.

%B Computer Analysis of Images and Patterns %S Lecture Notes in Computer Science %I Springer Verlag %C Berlin; Heidelberg %P 165 - 172 %8 Aug 2007 %@ 978-3-540-74271-5 %G eng %9 Conference paper %! LNCS %R 10.1007/978-3-540-74272-2_21 %0 Journal Article %J INVESTIGATIVE OPHTHALMOLOGY & VISUAL SCIENCE %D 2007 %T Extraction of an Age-Related Biomarker From Retinal Images Using Appearance Based Approach %A Georg Michelson %A Simone Wärntges %A Rudriger Bock %A László Gábor Nyúl %A Joachim Hornegger %X

Purpose:To develop an appropriate algorithm from retina imagesusing an appearance-based version of the Principal Component Analysis and to test the age-related biomarker’s significance for patients at cardiovascular risk. Methods:Sixty-five men (age, 44.2 ± 11.4 years) and 60 women (age, 48.8 ± 12.6 years) without cardiovascular risk factors and without pathologic eye diagnosis were acquired during a clinical non-experimental cross-sectional survey and represented the control group. Forty-four hypertensive men (age, 45.5 ± 9.4 years; hypertensive for 5.9 ± 6.7 years) and 26 hypertensive women (age, 51.2 ± 7.3 years; hypertensive for 7.9 ± 7.1 years) as well as 57 male smokers (age, 41.8 ± 8.5 years; smoking for 20.6 ± 9.8 years; 15.3 ± 8.6 cigarettes per day) and 60 female smokers (age, 43.2 ± 9.5 years; smoking for 20.1 ± 10.7 years; 13.5 ± 8.1 cigarettes per day) were matched for age and sex to the respective number of control subjects. Results:The reliability of the algorithm was 0.958. The retinal biomarker correlated with age (men, -0.284, p = 0.017; women, -0.374, p = 0.001). Smokers showed a lower biomarker value (male, -0.16 ± 1.29; female, -0.12 ± 0.11) than age-matched control subjects (male, 0.72 ± 0.92, p < 0.001; female, 0.24 ± 0.98, p = 0.048). Hypertension had a similar influence to the biomarker in men (0.10 ± 0.84), but not in women (-0.46 ± 1.23) as compared to age-matched controls (male, 0.57 ± 0.95, p = 0.01; female, 0.06 ± 0.99, p = 0.09). Conclusions:The algorithm of the appearance-based version of the Principal Component Analysis identified an age-related image feature dependent on light intensity with a strong influence to the temporal parapapillary area. It may be used to identify patients at cardiovascular risk.

%B INVESTIGATIVE OPHTHALMOLOGY & VISUAL SCIENCE %V 48 %P 2167 %8 2007 %@ 0146-0404 %G eng %N 5 %9 Journal article %! INVEST OPHTH VIS SCI %0 Book Section %B 52nd IWK - Internationales Wissenschaftliches Kolloquium - Volume II. %D 2007 %T Eye Fundus Image Processing System for Automated Glaucoma Classification %A Jörg Meier %A Rudriger Bock %A László Gábor Nyúl %A Georg Michelson %E P Scharff %B 52nd IWK - Internationales Wissenschaftliches Kolloquium - Volume II. %I Technische Universitat %C Ilmenau %P 81 - 84 %8 Sep 2007 %G eng %U http://www.db-thueringen.de/servlets/DerivateServlet/Derivate-12272/IWK_2007_2.pdf %9 Conference paper %0 Conference Paper %B BMT 2007: 41. Jahrestagung der Deutschen Gesellschaft für Biomedizinische Technik im VDE %D 2007 %T Retina Image Analysis System for Glaucoma Detection %A Rudriger Bock %A Jörg Meier %A László Gábor Nyúl %A Georg Michelson %A Joachim Hornegger %B BMT 2007: 41. Jahrestagung der Deutschen Gesellschaft für Biomedizinische Technik im VDE %C Aachen, Germany %8 Sep 2007 %G eng %9 Conference paper