Swarm Intelligence Optimization-based Method For Optic Disc Segmentation

With the development of the times,the wide application of electronic screens,the excessive use of the eyes has become the norm.Among the many ophthalmic diseases,glaucoma,as the largest source of blindness in addition to cataracts,is not only difficult to diagnose,but also leads to physiological lesions in fundus structures,resulting in permanent blindness.At present,many scholars use computer-aided systems to assist in the diagnosis of glaucoma.The effective division of the cup area on the fundus map has extraordinary significance for the diagnosis of glaucoma,which is also the main research direction of this paper.In the process of finding the relevant data set,we found that the data set standard of the fundus map is not uniform,and contains a lot of redundant information,which requires pre-processing of the original image.First of all,the grayscale operation,taking the red channel with the most obvious outline of the disc,and then using Lanczos interpolation to reduce the size of the picture to a unified standard.In this paper,the advantages and disadvantages of the existing fundus segmentation algorithm are compared through research and analysis,and the random particle swarm algorithm is selected for the optic disc segmentation in view of the phenomenon that the number of fundus diagram data is small and the distribution is uneven.The particle swarm algorithm is not only simple in parameters and easy to implement,but also has strong stability,the data requirements are not high,and there is no need for time-consuming pre-training sessions.The original image is preprocessed before the segmentation work is performed.For the field of optic disc segmentation problem of the algorithm of random particle swarm,this paper mainly explores two aspects:1.The optic disc on the fundus diagram is a smooth approximate elliptic curve,based on prior knowledge,this paper proposes a multi-population Particle Swarm Optimization based on active shape model(ASMPSO).This method draws on the idea of active shape model,transforms the optic disc segmentation problem into a set of extreme value problems,and realizes the segmentation algorithm by solving the method of constructing contour lines at multiple optimal positions.A variety of swarm random particle swarm algorithm is selected,the population is divided into multiple subpopulations,the subgroups are independent of each other,no information sharing is carried out,and the particles in the subpopulation are searched and optimized according to the traditional random particle swarm method,and the gradient is used to calculate the suitability of particles and the fitness of the contour line,and the optimal solution is required through multiple iterations.The effect of different subpopulations on the classic public dataset Drishti-GS dataset explores the effect of disc segmentation on different sub-populations,and the Drishti-GS and Rim-ONE V3 are selected,which are compared with four optic disc segmentation methods based on superpixel,activate contour model,threshold and active shape model,which proves the effectiveness of the segmentation algorithm,but the ASMPSO algorithm also has certain deficiencies,and the solved optic disc contour is not smooth enough,and there are some burrs.2.To compensate for the shortcomings of the ASMPSO algorithm,this paper proposes a Multi Population Particle Swarm Optimization based on Elastic net method(ENPSO).Based on the ASMPSO method,the modeled contour line is changed to an elastic contour line,and the particles become elastic particles.During particle flight,they are additionally subjected to the gravitational pull of adjacent elastic particles on the elastic contour line,in this way to achieve information sharing between sub-populations of particles.Based on Drishti-GS,the effects of the three weights of individual learning factor,subpopulation learning factor and elastic particle learning factor on particle search ability were analyzed and compared with THE ASMPSO algorithm.Experimental results show that the proposed ENPSO has higher stability and robustness.