Retinal vessel segmentation using multi-scale textons derived from keypoints

Abstract

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.

Recent Development on Detection Methods for the Diagnosis of Diabetic Retinopathy

“…Their framework reported an accuracy of 95.36% at the cost of a long processing time and low specificity (SP) for vessel segmentation. Zhang et al [48] proposed an unsupervised vessel segmentation approach in which labelling inconsistencies from ground truth labels were optimized. Key point descriptors were furnished to generate a texton dictionary to distinguish vessel pixels from on-vessel intensities.…”

Multilevel and Multiscale Deep Neural Network for Retinal Blood Vessel Segmentation

“…Matched filter kernels respond to both the vessel and non-vessel structures thereby reducing their performance [17,18]. Although multiscale approaches selectively use multiple scales to extract all the useful vessel information, they could not improve the accuracy in vessel segmentation [22,23]. Region growing method has improved AUC values such as 0.967 for both DRIVE and STARE because of the novel stopping criteria [26].…”

“…Matched filtering approaches are unable to segment vessels at the region of pathologies, central reflex and low contrast [15][16][17][18][19]. Information obtained from multiple scales was able to segment vessels but it could not detect low contrast thin vessels [20][21][22][23]. Region growing methods also proved to be a useful technique in vessel segmentation but expertise is needed in vessel seed point setting and in the formulation of a stopping rule [24][25][26].…”

Integrated Gabor Filter and Trilateral Filter for Exudate Extraction in Fundus Images

“…[2015] [6] presented an algorithm for segmenting retinal vessel structures which makes use of a texton dictionary to categorize vessel pixels or non-vessel pixels. This algorithm also uses Key points which are a small set of image features to derive the parameters for filtering.…”