Research On Medical Image Segmentation Method Based On Improved Meta-heuristic Algorithm

With the in-depth development of theories and technologies related to artificial intelligence(AI),it has been widely used in various fields.In the medical field,intelligent medical diagnosis based on AI technology has become a research hotspot.In medical diagnosis data,the results of pathology image processing are an important basis for disease diagnosis,and pathology image segmentation is a key technology for medical image processing.Among them,the image segmentation method based on thresholding,which has the advantages of simple implementation,strong generalization and high segmentation efficiency,has become an important means of medical image segmentation.However,because the traditional threshold segmentation method uses the exhaustive method to search for thresholds,which leads to an exponential increase in computation with the increase of the threshold dimension.On the other hand,due to the refinement and expansion of the color gamut of medical images,the traditional discrete threshold segmentation methods can hardly achieve the requirements of medical image refinement segmentation.The metaheuristic algorithm can avoid such problems because of its own characteristics,but due to the complexity of the actual problem,there are defects such as poor global search ability,low convergence accuracy,and easy to fall into local optimality in the solution process,which will lead to the degradation of the performance of the threshold segmentation method and affect the final effect of image segmentation.To address the above problems,this paper proposes three image segmentation models based on three metaheuristic optimization algorithms(artificial bee colony optimization algorithm,multi-verse optimization algorithm and bat optimization algorithm),aiming to achieve accurate segmentation of images and thus provide effective tools for clinical diagnosis.Firstly,considering the limitations of the three metaheuristic algorithms,including global search capability,local exploitation capability,convergence accuracy,and ability to jump out of local optimum,three improved algorithms are proposed,namely,the artificial bee colony algorithm based on the vertical and horizontal crossover mechanism,the multiverse algorithm based on the vertical and horizontal crossover mechanism,and the bat algorithm based on the hybrid strategy.Further,based on the above improved algorithms combined with the non-local mean filter denoising method,2D Kapur entropy image segmentation method,and 2D histogram information characterization method,a multi-threshold image segmentation model based on the improved artificial bee colony algorithm,a multi-threshold image segmentation model based on the improved multiverse algorithm,and a multi-threshold image segmentation model based on the improved bat algorithm are proposed.The experimental results show that the proposed segmentation models have greater advantages in segmentation accuracy and segmentation speed compared with the same type of methods,and can effectively achieve the pathological image segmentation of COVID-19 and thyroid cancer.The details are as follows.(1)To effectively denoise and segment COVID-19 images,this paper improves the artificial swarm algorithm and combines it with image segmentation methods to construct an artificial swarm multi-threshold image segmentation model based on vertical and horizontal crossover.First,to address the problems of narrow search range and slow search efficiency of the original artificial bee colony algorithm in the presearch process,the longitudinal crossover mechanism is introduced into the pre iteration of the algorithm,and an improved artificial bee colony optimization algorithm based on re-search is proposed.Further,comparative simulation experiments are designed relying on the IEEE CEC2014 test set,and the results show that the proposed improved algorithm has stronger search capability compared with the original algorithm.Compared with 12 algorithms of the same type,the convergence speed and convergence accuracy of the proposed algorithm in the iterative process have greater advantages.Finally,the improved artificial bee colony algorithm is used to optimize the thresholding search process of Kapur entropy,and combined with image denoising methods and image information characterization methods,a multi-threshold segmentation model is proposed for COVID-19 pathology images,and the experimental results show that the model can effectively segment pathology images.(2)Since COVID-19 pathology images are characterized by multiple sources of data and complicated subdivision types,a separate image segmentation model is not sufficient to effectively handle all types of COVID-19 images,and an image segmentation model based on an improved multiverse optimization algorithm is proposed in this paper.First,to address the problem of unbalanced search performance and exploitation performance of the original multiverse optimization algorithm in the process of finding the best,the vertical and horizontal crossover mechanism is introduced into the global iteration of the algorithm,and an improved multiverse optimization algorithm with balanced search and exploitation is proposed.Similarly,comparative simulation experiments are designed based on the IEEE CEC2014 test set,and the experimental results show that the proposed improved algorithm has a wider search range and higher exploitation accuracy compared to the original algorithm.The experiments also compare the improved algorithm with four peer algorithms and five variant algorithms,and the results show that the proposed algorithm has stronger search efficiency and convergence accuracy.Finally,the improved multiverse algorithm was used to optimize the threshold search process of Kapur entropy and combined with the image segmentation method to propose a multi-threshold segmentation model for COVID-19 pathology images,and the experimental results showed that the model can better determine the classification based on the key threshold and obtain excellent image segmentation results.(3)To explore the generalization ability and migration ability of the proposed image segmentation model,we aim to solve the problem of segmenting different disease images.In this paper,an image segmentation model based on an improved bat optimization algorithm is proposed and successfully applied to image segmentation of thyroid cancer.First,to address the problem that the original bat optimization algorithm has a narrow coverage of solution individuals during initialization,the Lévy flight mechanism replaces the original random initialization operation,which achieves a lowsequence initialization of the algorithm and facilitates further search of the algorithm.On the other hand,to address the problem that the bat algorithm easily falls into local optimum,the guided crossover mechanism is used for the communication between bat individuals,which enhances the update frequency and update effect of the search individual positions,and improves the ability of the bat optimization algorithm to avoid falling into local optimum and its ability to obtain high-quality solutions.The results of the comparison with the original algorithm in the IEEE CEC2014 benchmark function experiments confirm the correctness and effectiveness of the improved idea.Similarly,the improved bat algorithm is compared with eight algorithms of the same type,and the experimental results show that the improved algorithm has better global search capability and higher optimization accuracy.Finally,the improved bat optimization algorithm is combined with image denoising and segmentation methods to propose a multi-threshold segmentation model for thyroid cancer pathology images,and the experimental results show that the proposed model has higher segmentation accuracy and segmentation efficiency compared with the same type of algorithms when dealing with complex thyroid cancer images,and can effectively denoise and segment thyroid cancer images.