A high accuracy segmentation method for retinal blood vessel detection based on hybrid filters and an adaptive thresholding

The field of computer-assisted retinal vascular segmentation is crucial since it aids in the diagnosis of disorders like diabetic retinopathy. The segmentation of retinal images deals with several problems, including the appearance of pseudo vascularization, difficulty in detecting thin vessels, and the enhancement of low-resolution images. This study aims to propose a new unsupervised method for retinal vascular segmentation that ensures high accuracy detection comparatively with previous studies. The proposed method can achieve good performance without prior training or turning. It is based on an efficient hybrid combination of many well-known intensity transformations and filters, followed by an adaptive thresholding algorithm. Firstly, contrast limited adaptive histogram equalization (CLAHE) and bottom-hat (BTH) filtering are applied to increase the contrast between the vascular and the fundus. To bring up the vessel tree structure against a non-uniform image background, a Jerman filtering is performed. Then, reconstruction processes, bowler-hat (BLH) filtering, and the generated field of view (FoV) mask are applied to preserve image details and remove any noise. Finally, an adaptive threshold is used to classify vessel and non-vessel pixels. The impact of the proposed segmentation model has been evaluated on the open-access STARE and DRIVE databases, reaching an accuracy index of 0.9618 and 0.9586, and a specificity index of 0.9810 and 0.9874, respectively. The suggested segmentation method proved more accurate and more efficient than the results of some other current methods. © The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2022.