Research On Salp Swarm Algorithm And Its Application In Medical Diagnosis

Swarm intelligence optimization method is playing an increasingly important role in r eal life.salp Swarm algorithm(SSA)is a newly proposed algorithm,which is a typical repr esentative of swarm intelligence optimization algorithm.It is inspired by the behavior of sal ps,which use the movement and interaction of its chains as individual movement rules to u pdate the individual position and gradually approach the food source.In many practical appl ication problems,SSA shows good performance due to its few parameters,excellent perform ance and the ability of crossing multi-dimensional feature space and aiming at food in all d irections,reducing the probability of falling into local optimum to a certain extent.However,with the application of SSA becoming increasingly prevalent,its shortcomings are gradually exposed,especially in solving the problem of medical diagnosis,meaning that the optimizat ion ability of SSA needs to be further improved.In order to improve the optimization abilit y of SSA in medical diagnosis problems,this study mainly aims at the shortcomings of SS A,and introduces a variety of new mechanisms to further enhance the performance of SSA algorithm,which can help it find a better balance between global optimization and local o ptimization.The main work of this paper includes the following aspects:1.In this paper,the research status of optimization algorithm and medical diagnosis of salp swarm algorithm is presented.At the same time,the existing problems in these fields are summarized and analyzed,and its development trend and problems are pointed out.In addition,some methods are introduced,such as machine learning,feature selection and image segmentation,etc.The paper mainly discusses and analyzes the fuzzy k-nearest neighbor(FKNN)algorithm,SSA and other related theories,the existing problems and the improvement direction.2.In order to achieve better classification performance in medical diagnosis,this paper uses integrated mutation strategy(CMS)to strengthen the development and exploration ability of SSA based on the shortcomings of SSA,so that it can obtain better convergence precision when executing the integrated mutation strategy in parallel.At the same time,the restart mechanism(RS)is utilized to help the salp to restart the initialization strategy when its position is not improved after a number of iterations,as so to prevent the whole population from falling into the local optimal state.The improved SSA method is used to optimize FKNN model,and construct the CMSRSSSA-FKNN prediction model possessing the best performance.The model has been verified on some diseases such as breast cancer,liver disease,heart disease and diabetes,and the experimental results reveal that compared with other machine learning methods,CMSRSSSA-FKNN model has achievedsignificant improvement in four evaluation indicators.3.In order to enhance the effect of feature selection of medical data,this paper introduces quasi opposition-based learning strategy and bare bone mechanism in SSA,and a quasi opposition-based learning(QBSSA)algorithm based on bare bone mechanism is proposed.It shows better performance in both standard test set and feature selection of actual medical data.Compared with 10 improved swarm intelligence algorithms on CEC2017 competition data set,the results show that the proposed QBSSA has better results in terms of solution quality and convergence.QBSSA is superior to the other five algorithms in three different dimensions(fitness,error value and feature number)in the selection of key pathogenic features of breast cancer and lymphography.4.In order to improve the effect of medical image segmentation,fractal search mechanism and Gaussian skeleton mechanism are introduced into the original SSA,and an SSA algorithm based on skeleton fractal mechanism(GBSFSSSA)is proposed.The optimal threshold in multi threshold image segmentation is searched by non-local mean and Kapur entropy.Excellent results have been achieved in the segmentation of lupus erythematosus renal pathological images.In the experiment,the GBSFSSSA proposed in this paper is compared with five other algorithms,showing that the overall performance of GBSFSSSA proposed in this paper is better than other contrast algorithms in three different dimensions(PSNR,SSIM and FSIM).