The retinal vein segmentation assumes a significant job in programmed or PC helped diagnosis of retinopathy. Manual vein segmentation is very tedious and requires a lot of subject specific information. Also, the veins are just a couple of pixels wide and spread across the whole fundus image. This further impedes the ongoing frameworks from mechanizing the retinal vein segmentation productively. In this paper, we propose a novel Hybrid Differential Evolution (DE) to complete programmed retinal vein segmentation. The proposed DE starts by generating three dynamic sets of populations focusing on thick veins, thin veins and non-veins. It also employs a non-replaceable memory concept to store all the three initial population members. Then in order to further balance the local and global exploration a micro firefly optimization algorithm with three mutations techniques is embedded in the core of DE. Various classifiers, for example, Neural Networks (NN), Support vector machines (SVM), NN and SVM based ensemble are utilized to additionally evaluate the performance of the segmentation system. The proposed system is assessed on the openly accessible DRIVE, STARE and HRF retinal image datasets and it outflanked cutting edge strategies with a high normal precision of 98.5% alongside high affectability and explicitness.
|Title of host publication||Recent Advances in AI-enabled Automated Medical Diagnosis|
|Editors||Richard Jiang, Danny Crookes, Hua-Liang Wei, Li Zhang, Paul Chazot|
|Publication status||Published - 20 Oct 2022|