Retinal vessel segmentation using multiwavelet kernels and multiscale hierarchical decomposition

Wang, Yangfan; Ji, Guangrong; Lin, Ping; Trucco, Emanuele

doi:10.1016/j.patcog.2012.12.014

Cited by 139 publications

(78 citation statements)

References 39 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This ridge detection method is based on the retinal vessel segmentation method [12][13][14]18]. Let us denote ( ) with = ( , ) to be the SW image, where the first derivative of the image ( ) in the direction perpendicular to the ridge tangent has a zerocrossing.…”

Section: Ridge Detectionmentioning

confidence: 99%

Feature Extraction and Classification to Diagnose Hypoxic-Ischemic Encephalopathy Patients by Using Susceptibility-Weighted MRI Images

Mahmoodi

Darekar

et al. 2017

Communications in Computer and Information Science

View full text Add to dashboard Cite

Abstract. In this paper. a method is presented to enable automatic classification of the degree of abnormality of susceptibility-weighted images (SWI) acquired from babies with hypoxic-ischemic encephalopathy (HIE), in order to more accurately predict eventual cognitive and motor outcomes in these infants. SWI images highlight the cerebral venous vasculature and can reflect abnormalities in blood flow and oxygenation, which may be linked to adverse outcomes. A qualitative score based on magnetic resonance imaging (MRI) analyses is assigned to SWIs by specialists to determine the severity of abnormality in an HIE patient. The method allows the detection of image ridges, representing the vessels in SWIs, and the histogram of the ridges grey scales. A curve with only four parameters is fitted to the histograms. These parameters are then used to estimate the SWI abnormality score. The images are classified by using a kNNand multiple SVM classifiers based on the parameters of the fitting curves. The algorithm is tested on an SWI-MRI dataset consisting of 10 healthy infants and 48 infants with HIE with a range of SWI abnormality scores between 1 and 7. The accuracy of classifying babies with HIE vs. those without (ie: healthy controls) using our algorithm with a leave-one-out strategy is measured as 91.38%. Our method is fast and could increase the prognostic value of these scans, thereby improving management of the condition, as well as elucidating the disease mechanisms of HIE.

show abstract

Section: Ridge Detectionmentioning

confidence: 99%

Feature Extraction and Classification to Diagnose Hypoxic-Ischemic Encephalopathy Patients by Using Susceptibility-Weighted MRI Images

Mahmoodi

Darekar

et al. 2017

Communications in Computer and Information Science

View full text Add to dashboard Cite

show abstract

“…Although supervised segmentation methods (reviewed in section 2) are more competitive in terms of performance than unsupervised approaches [36] [37] [38], their dependence on ground-truth and the problem of intra-and inter-observer variability limits their robustness [34]. In this and many other fields of medical image understanding, ground truth tends to be sparse as the laborious and skilled nature of the task makes it very expensive to acquire.…”

Section: Of 33mentioning

confidence: 99%

“…Wang et al [38] proposed an unsupervised retinal vessel segmentation method that does not require a pre-processing stage. The vessels are initially enhanced using a matched filter with a multi-wavelet kernel which is capable of responding to blood vessels and non-vessel edges.…”

Section: Related Workmentioning

confidence: 99%

Retinal vessel segmentation using multi-scale textons derived from keypoints

Zhang

Fisher

Wang

2015

Computerized Medical Imaging and Graphics

View full text Add to dashboard Cite

Highlights We present a retinal vessel segmentation approach that uses textons. Vessel textons are derived from responses of a multi-scale Gabor filter bank. Page 2 of 33A c c e p t e d M a n u s c r i p tRetinal vessel segmentation using multi-scale textons derived from keypoints 2  We train on keypoint descriptors instead of labelled ground truth. We show our unsupervised approach performs well compared to previous work. Our method outperforms other unsupervised approaches on the Drive data set. AbstractThis paper presents a retinal vessel segmentation algorithm which uses a texton dictionary to classify vessel/non-vessel pixels. However, in contrast to previous work where filter parameters are learnt from manually labelled image pixels our filter parameters are derived from a smaller set of image features that we call keypoints. A Gabor filter bank, parameterised empirically by ROC analysis, is used to extract keypoints representing significant scale specific vessel features using an approach inspired by the SIFT algorithm. We first determine keypoints using a validation set and then derive seeds from these points to initialise a k-means clustering algorithm which builds a texton dictionary from another training set. During testing we use a simple 1-NN classifier to identify vessel/non-vessel pixels and evaluate our system using the DRIVE database. We achieve average values of sensitivity, specificity and accuracy of 78.12%, 96.68% and 95.05% respectively. We find that clusters of filter responses from keypoints are more robust than those derived from hand-labelled pixels. This, in turn yields textons more representative of vessel/non-vessel classes and mitigates problems arising due to intra and inter-observer variability.

show abstract

“…The training data or hand labelled ground truths are not needed for the design of algorithm in these approaches. These approaches are matched filter-based (Chaudhuri et al;1989;Hoover et al;Chanwimaluang and Fan;Cinsdikici and Aydın;2009;Zhang et al;Chakraborti et al;, scale space-based (Martínez-Pérez et al;Vlachos and Dermatas;Wang et al;, tracking-based , model-based (Szpak and Tapamo;Xiao et al;, adaptive thresholding-based (Jiang and Mojon;Cornforth et al;2005;Akram and Khan;, mathematical morphology-based (Zana and Klein;Mendonca and Campilho;Jiménez et al;Miri and Mahloojifar;2012a) and clustering-based (Tolias and Panas;1998;Yang et al;Kande et al;2011a,b;Sun et al;Saffarzadeh et al;. Chaudhuri et al (1989) implemented matched filter response (MFR) by initially approximating the intensity of gray-level profiles of the cross-sections of retinal vessels using a Gaussian shaped curve.…”

Section: Introductionmentioning

confidence: 99%

“…Vlachos and Dermatas (2010) proposed a retinal vessel segmentation method based on multi-scale line-tracking procedure and morphological post-processing. Wang et al (2013) proposed multi-wavelet kernels and multi-scale hierarchical decomposition. Vessels were enhanced using matched filtering with multi-wavelet kernels.…”

Section: Introductionmentioning

confidence: 99%

Automatic Retinal Vessel Detection and Tortuosity Measurement

Mapayi¹,

Tapamo²,

Viriri³

et al. 2016

Image Anal Stereol

View full text Add to dashboard Cite

As retinopathies continue to be major causes of visual loss and blindness worldwide, early detection and management of these diseases will help achieve significant reduction of blindness cases. However, an efficient automatic retinal vessel segmentation approach remains a challenge. Since efficient vessel network detection is a very important step needed in ophthalmology for reliable retinal vessel characterization, this paper presents the study on the combination of difference image and k-means clustering for the segmentation of retinal vessels. Stationary points in the vessel center-lines are used to model the detection of twists in the vessel segments. The combination of arc-chord ratio with stationary points is used to compute tortuosity index. Experimental results show that the proposed k-means combined with difference image achieved a robust segmentation of retinal vessels. A maximum average accuracy of 0.9556 and a maximum average sensitivity of 0.7581 were achieved on DRIVE database while a maximum average accuracy of 0.9509 and a maximum average sensitivity of 0.7666 were achieved on STARE database. When compared with the previously proposed techniques on DRIVE and STARE databases, the proposed technique yields higher mean sensitivity and mean accuracy rates in the same range of very good specificity. In a related development, a non-normalized tortuosity index that combined distance metric and the vessel twist frequency proposed in this paper also achieved a strong correlation of 0.80 with the expert ground truth.

show abstract

Retinal vessel segmentation using multiwavelet kernels and multiscale hierarchical decomposition

Cited by 139 publications

References 39 publications

Feature Extraction and Classification to Diagnose Hypoxic-Ischemic Encephalopathy Patients by Using Susceptibility-Weighted MRI Images

Feature Extraction and Classification to Diagnose Hypoxic-Ischemic Encephalopathy Patients by Using Susceptibility-Weighted MRI Images

Retinal vessel segmentation using multi-scale textons derived from keypoints

Automatic Retinal Vessel Detection and Tortuosity Measurement

Contact Info

Product

Resources

About