Lucia Ballerini

Leveraging Multiscale Hessian-Based Enhancement With a Novel Exudate Inpainting Technique for Retinal Vessel Segmentation

Accurate vessel detection in retinal images is an important and difficult task. Detection is made more challenging in pathological images with the presence of exudates and other abnormalities. In this paper, we present a new unsupervised vessel segmentation approach to address this problem. A novel inpainting filter, called neighborhood estimator before filling, is proposed to inpaint exudates in a way that nearby false positives are significantly reduced during vessel enhancement. Retinal vascular enhancement is achieved with a multiple-scale Hessian approach. Experimental results show that the proposed vessel segmentation method outperforms state-of-the-art algorithms reported in the recent literature, both visually and in terms of quantitative measurements, with overall mean accuracy of 95.62% on the STARE dataset and 95.81% on the HRF dataset.

Retinal vessel classification: Sorting arteries and veins

For the discovery of biomarkers in the retinal vasculature it is essential to classify vessels into arteries and veins. We automatically classify retinal vessels as arteries or veins based on colour features using a Gaussian Mixture Model, an Expectation-Maximization (GMM-EM) unsupervised classifier, and a quadrant-pairwise approach. Classification is performed on illumination-corrected images. 406 vessels from 35 images were processed resulting in 92% correct classification (when unlabelled vessels are not taken into account) as compared to 87.6%, 90.08%, and 88.28% reported in [12] [14] and [15]. The classifier results were compared against two trained human graders to establish performance parameters to validate the success of classification method. The proposed system results in specificity of (0.8978, 0.9591) and precision (positive predicted value) of (0.9045, 0.9408) as compared to specificity of (0.8920, 0.7918) and precision of (0.8802, 0.8118) for (arteries, veins) respectively as reported in [13]. The classification accuracy was found to be 0.8719 and 0.8547 for veins and arteries, respectively.

Accurate and reliable segmentation of the optic disc in digital fundus images

Abstract. We describe a complete pipeline for the detection and accurate automatic segmentation of the optic disc in digital fundus images. This procedure provides separation of vascular information and accurate inpainting of vessel-removed images, symmetry-based optic disc localization, and fitting of incrementally complex contour models at increasing resolutions using information related to inpainted images and vessel masks. Validation experiments, performed on a large dataset of images of healthy and pathological eyes, annotated by experts and partially graded with a quality label, demonstrate the good performances of the proposed approach. The method is able to detect the optic disc and trace its contours better than the other systems presented in the literature and tested on the same data. The average error in the obtained contour masks is reasonably close to the interoperator errors and suitable for practical applications. The optic disc segmentation pipeline is currently integrated in a complete software suite for the semiautomatic quantification of retinal vessel properties from fundus camera images (VAMPIRE).

The use of radial symmetry to localize retinal landmarks

Locating the optic disc center and the fovea in digital fundus images is surprisingly difficult due to the variation range in color and contrast and the possible presence of pathologies creating bright spots or changing the appearance of retinal landmarks. These reasons make it difficult to find good templates of optic disc and fovea shape and color for pattern matching. In this paper we propose radial symmetry as the principal cue to locate both optic disc and macula centers. Centers of bright and dark circularly symmetrical regions with arbitrary radii, can be found robustly against changes in brightness and contrast by using the Fast Radial Symmetry transform. Detectors based on this transform coupled with a weak hypothesis on vessel density (optic disc intersects large vessels while the fovea lies in an avascular region), can provide a fast location of both OD and macula with accuracy similar or better than state-of-the-art methods. The approach has been chosen as the default technique for fast localization of the two landmarks in the VAMPIRE software suite.

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 retinal vessel classification using a Least Square-Support Vector Machine in VAMPIRE

It is important to classify retinal blood vessels into arterioles and venules for computerised analysis of the vasculature and to aid discovery of disease biomarkers. For instance, zone B is the standardised region of a retinal image utilised for the measurement of the arteriole to venule width ratio (AVR), a parameter indicative of microvascular health and systemic disease. We introduce a Least Square-Support Vector Machine (LS-SVM) classifier for the first time (to the best of our knowledge) to label automatically arterioles and venules. We use only 4 image features and consider vessels inside zone B (802 vessels from 70 fundus camera images) and in an extended zone (1,207 vessels, 70 fundus camera images). We achieve an accuracy of 94.88% and 93.96% in zone B and the extended zone, respectively, with a training set of 10 images and a testing set of 60 images. With a smaller training set of only 5 images and the same testing set we achieve an accuracy of 94.16% and 93.95%, respectively. This experiment was repeated five times by randomly choosing 10 and 5 images for the training set. Mean classification accuracy are close to the above mentioned result. We conclude that the performance of our system is very promising and outperforms most recently reported systems. Our approach requires smaller training data sets compared to others but still results in a similar or higher classification rate.

Spline-based refinement of vessel contours in fundus retinal images for width estimation

This paper presents a novel algorithm for refining the vessel contours obtained from retinal binary vessel maps. The refinement improves vessel width estimation. Our approach is based on fitting the two contours of each vessel in the binary map with a cubic spline curve, introducing a parallelism constraint between the two splines. The algorithm, evaluated on the REVIEW database, has a comparable accuracy to that of specialized, sophisticated width estimation algorithms.

Retinal Vessel Classification Based on Maximization of Squared-Loss Mutual Information

The classification of retinal vessels into arterioles and venules is important for any automated system for the detection of vascular changes in the retina and for the discovery of biomarkers associated with systemic diseases such as diabetes, hypertension, and cardiovascular disease. We introduce Squared-loss Mutual Information clustering (SMIC) for classifying arterioles and venules in retinal images for the first time (to the best of our knowledge). We classified vessels from 70 fundus camera images using only 4 colour features in zone B (802 vessels) and in an extended zone (1,207 vessels). We achieved an accuracy of 90.67 and 87.66 % in zone B and the extended zone, respectively. We further validated our algorithm by classifying vessels in zone B from two publically available datasets—INSPIRE-AVR (483 vessels from 40 images) and DRIVE (171 vessels from 20 test images). The classification rates obtained on INSPIRE-AVR and DRIVE dataset were 87.6 and 86.2 %, respectively. We also present a technique to sort the unclassified vessels which remained unlabeled by the SMIC algorithm.

Association between retinal vasculature and muscle mass in older people.

UNLABELLEDnSarcopenia in older people is a major health issue and its early detection could help target interventions and improve health. Evidence suggests that poor muscle mass is associated with greater arterial stiffness and cardiovascular risk. Arterial stiffness in turn is associated with smaller retinal artery width. This study examined the association of muscle mass in older people with retinal vascular width, a non-invasive measure of vascular function.nnnMETHODSnParticipants >65 years were recruited to a cross-sectional study.nnnEXCLUSIONSnInability to walk independently; diabetes mellitus; stroke (within 6 months), severe macular degeneration, glaucoma, retinal dystrophy; advanced cataract. Digital Retinal images of both eyes were analysed using the VAMPIRE software suite. Central Retinal Artery and Vein Equivalents (CRVE and CRAE) were measured. Body composition was measured using Dual Energy X ray Absorptimetry (DXA). Appendicular Skeletal Muscle Mass/Height(2) was calculated. Physical function was measured: 6-min walk distance, Short Physical performance battery, handgrip strength and quadriceps strength.nnnRESULTSn79 participants with mean age 72 (SD 6) years were recruited. 44% were female. Digital Retinal images of sufficient quality for measuring CRAE and CRVE were available for 51/75 (68%) of participants. Regression analysis showed significant association between larger ASMM/H(2) and smaller CRAE (β=-0.20, p=0.001) and CRVE (β=-0.12, p=0.05). Handgrip strength, body mass index and sex combined with CRAE explained 88% and with CRVE explained 86% of the variance in ASMM/H(2).nnnCONCLUSIONnLarger muscle mass was significantly associated with smaller retinal artery size in older people. This unexpected finding needs further investigation.

Retinal biomarker discovery for dementia in an elderly diabetic population

Dementia is a devastating disease, and has severe implications on affected individuals, their family and wider society. A growing body of literature is studying the association of retinal microvasculature measurement with dementia. We present a pilot study testing the strength of groups of conventional (semantic) and texture-based (non-semantic) measurements extracted from retinal fundus camera images to classify patients with and without dementia. We performed a 500-trial bootstrap analysis with regularized logistic regression on a cohort of 1,742 elderly diabetic individuals (median age 72.2). Age was the strongest predictor for this elderly cohort. Semantic retinal measurements featured in up to 81% of the bootstrap trials, with arterial caliber and optic disk size chosen most often, suggesting that they do complement age when selected together in a classifier. Textural features were able to train classifiers that match the performance of age, suggesting they are potentially a rich source of information for dementia outcome classification.