Carmen Alina Lupascu

FABC: Retinal Vessel Segmentation Using AdaBoost

This paper presents a method for automated vessel segmentation in retinal images. For each pixel in the field of view of the image, a 41-D feature vector is constructed, encoding information on the local intensity structure, spatial properties, and geometry at multiple scales. An AdaBoost classifier is trained on 789 914 gold standard examples of vessel and nonvessel pixels, then used for classifying previously unseen images. The algorithm was tested on the public digital retinal images for vessel extraction (DRIVE) set, frequently used in the literature and consisting of 40 manually labeled images with gold standard. Results were compared experimentally with those of eight algorithms as well as the additional manual segmentation provided by DRIVE. Training was conducted confined to the dedicated training set from the DRIVE database, and feature-based AdaBoost classifier (FABC) was tested on the 20 images from the test set. FABC achieved an area under the receiver operating characteristic (ROC) curve of 0.9561, in line with state-of-the-art approaches, but outperforming their accuracy (0.9597 versus 0.9473 for the nearest performer).

VAMPIRE: Vessel assessment and measurement platform for images of the REtina

We present VAMPIRE, a software application for efficient, semi-automatic quantification of retinal vessel properties with large collections of fundus camera images. VAMPIRE is also an international collaborative project of four image processing groups and five clinical centres. The system provides automatic detection of retinal landmarks (optic disc, vasculature), and quantifies key parameters used frequently in investigative studies: vessel width, vessel branching coefficients, and tortuosity. The ultimate vision is to make VAMPIRE available as a public tool, to support quantification and analysis of large collections of fundus camera images.

Automated Detection of Optic Disc Location in Retinal Images

This contribution presents an automated method to locate the optic disc in color fundus images. The method uses texture descriptors and a regression based method in order to determine the best circle that fits the optic disc. The best circle is chosen from a set of circles determined with an innovative method, not using the Hough transform as past approaches. An evaluation of the proposed method has been done using a database of 40 images. On this data set, our method achieved 95% success rate for the localization of the optic disc and 70% success rate for the identification of the optic disc contour (as a circle).

Accurate estimation of retinal vessel width using bagged decision trees and an extended multiresolution Hermite model

We present an algorithm estimating the width of retinal vessels in fundus camera images. The algorithm uses a novel parametric surface model of the cross-sectional intensities of vessels, and ensembles of bagged decision trees to estimate the local width from the parameters of the best-fit surface. We report comparative tests with REVIEW, currently the public database of reference for retinal width estimation, containing 16 images with 193 annotated vessel segments and 5066 profile points annotated manually by three independent experts. Comparative tests are reported also with our own set of 378 vessel widths selected sparsely in 38 images from the Tayside Scotland diabetic retinopathy screening programme and annotated manually by two clinicians. We obtain considerably better accuracies compared to leading methods in REVIEW tests and in Tayside tests. An important advantage of our method is its stability (success rate, i.e., meaningful measurement returned, of 100% on all REVIEW data sets and on the Tayside data set) compared to a variety of methods from the literature. We also find that results depend crucially on testing data and conditions, and discuss criteria for selecting a training set yielding optimal accuracy.

Novel VAMPIRE algorithms for quantitative analysis of the retinal vasculature

This paper summarizes three recent, novel algorithms developed within VAMPIRE, namely optic disc and macula detection, arteryvein classification, and enhancement of binary vessel masks, and their performance assessment. VAMPIRE is an international collaboration growing a suite of software tools to allow efficient quantification of morphological properties of the retinal vasculature in large collections of fundus camera images. VAMPIRE measurements are currently mostly used in biomarker research, i.e., investigating associations between the morphology of the retinal vasculature and a number of clinical and cognitive conditions.

Automatic unsupervised segmentation of retinal vessels using self-organizing maps and K-means clustering

In this paper an automatic unsupervised method for the segmentation of retinal vessels is proposed. A Self-Organizing Map is trained on a portion of the same image that is tested and K-means clustering algorithm is used to divide the map units in 2 classes. The entire image is again input for the Self-Organizing Map, and the class of each pixel will be the class of the best matching unit on the Self-Organizing Map. Finally, the vessel network is post-processed using a hill climbing strategy on the connected components of the segmented image. The experimental evaluation on the publicly available DRIVE database shows accurate extraction of vessels network and a good agreement between our segmentation and the ground truth. The mean accuracy, 0.9459 with a standard deviation of 0.0094, is outperforming the manual segmentation rates obtained by other widely used unsupervised methods. A good kappa value of 0.6562 is inline with state-of-the-art supervised and unsupervised approaches.

A Comparative Study on Feature Selection for Retinal Vessel Segmentation Using FABC

This paper presents a comparative study on five feature selection heuristics applied to a retinal image database called DRIVE. Features are chosen from a feature vector (encoding local information, but as well information from structures and shapes available in the image) constructed for each pixel in the field of view (FOV) of the image. After selecting the most discriminatory features, an AdaBoost classifier is applied for training. The results of classifications are used to compare the effectiveness of the five feature selection methods.

Blood vessel segmentation and width estimation in ultra-wide field scanning laser ophthalmoscopy

Features of the retinal vasculature, such as vessel widths, are considered biomarkers for systemic disease. The aim of this work is to present a supervised approach to vessel segmentation in ultra-wide field of view scanning laser ophthalmoscope (UWFoV SLO) images and to evaluate its performance in terms of segmentation and vessel width estimation accuracy. The results of the proposed method are compared with ground truth measurements from human observers and with existing state-of-the-art techniques developed for fundus camera images that we optimized for UWFoV SLO images. Our algorithm is based on multi-scale matched filters, a neural network classifier and hysteresis thresholding. After spline-based refinement of the detected vessel contours, the vessel widths are estimated from the binary maps. Such analysis is performed on SLO images for the first time. The proposed method achieves the best results, both in vessel segmentation and in width estimation, in comparison to other automatic techniques.

Characterisation of human non-proliferative diabetic retinopathy using the fractal analysis.

AIM To investigate and quantify changes in the branching patterns of the retina vascular network in diabetes using the fractal analysis method. METHODS This was a clinic-based prospective study of 172 participants managed at the Ophthalmological Clinic of Cluj-Napoca, Romania, between January 2012 and December 2013. A set of 172 segmented and skeletonized human retinal images, corresponding to both normal (24 images) and pathological (148 images) states of the retina were examined. An automatic unsupervised method for retinal vessel segmentation was applied before fractal analysis. The fractal analyses of the retinal digital images were performed using the fractal analysis software ImageJ. Statistical analyses were performed for these groups using Microsoft Office Excel 2003 and GraphPad InStat software. RESULTS It was found that subtle changes in the vascular network geometry of the human retina are influenced by diabetic retinopathy (DR) and can be estimated using the fractal geometry. The average of fractal dimensions D for the normal images (segmented and skeletonized versions) is slightly lower than the corresponding values of mild non-proliferative DR (NPDR) images (segmented and skeletonized versions). The average of fractal dimensions D for the normal images (segmented and skeletonized versions) is higher than the corresponding values of moderate NPDR images (segmented and skeletonized versions). The lowest values were found for the corresponding values of severe NPDR images (segmented and skeletonized versions). CONCLUSION The fractal analysis of fundus photographs may be used for a more complete undeTrstanding of the early and basic pathophysiological mechanisms of diabetes. The architecture of the retinal microvasculature in diabetes can be quantitative quantified by means of the fractal dimension. Microvascular abnormalities on retinal imaging may elucidate early mechanistic pathways for microvascular complications and distinguish patients with DR from healthy individuals.

Stable automatic unsupervised segmentation of retinal vessels using self-organizing maps and a modified fuzzy C-means clustering

In this paper an automatic unsupervised method for the segmentation of retinal vessels is proposed. Three features are extracted from the tested image. The features are scaled down by a factor of 2 and mapped into a Self-Organizing Map. A modified Fuzzy C-Means clustering algorithm is used to divide the neuron units of the map in 2 classes. The entire image is again input for the Self-Organizing Map and the class of each pixel will be the class of its best matching unit in the Self-Organizing Map. Finally, the vessel network is post-processed using a hill climbing strategy on the connected components of the segmented image. The experimental evaluation on the DRIVE database shows accurate extraction of vessels network and a good agreement between our segmentation and the ground truth. The mean accuracy, 0.9482 with a standard deviation of 0.0075, is outperforming the manual segmentation rates obtained by other widely used unsupervised methods. A good kappa value of 0.6565 is comparable with state-of-the-art supervised or unsupervised approaches.