Optimizing The BP Network By Particle Swarm Optimization And Biogeography-Based Optimization Algorithms For Brain Image Detection

Computer-aided brain detection is a hot issue in the medical and computer fields.The solution of this problem has obvious beneficial effects on the prevention and treatment of brain diseases.This thesis takes magnetic resonance brain images as experimental subjects to study computer-aided medical detection methods that effectively classify healthy brain images and multiple disease brain images.The method is an intelligent classification method based on machine learning to improve the shortcomings of traditional classification techniques.The shortcomings include the problem that the manual classification has high time cost and large experience impact,and the problem that traditional neural network classification is easy to fall into local optimum and low performance.Firstly,the thesis obtains 61 human brain magnetic resonance images from the Harvard Medical School website.The dataset amplifies to 732 by data augmentation technology,and performs feature extraction and selection on the sample data.Then uses Discrete Wavelet Transform(DWT)and Wavelet Entropy(WE)to extract multi-scale features of brain images.Furthermore,seven features of the extracted brain images as input of BP neural network of the single hidden layer for training.Secondly,in order to avoid the network falling into the local optimal solution in the training process,the following three methods are included.The first is to propose Particle Swarm Optimization(PSO)algorithm to optimize the weight and bias of the network when considering its high precision,fast convergence,and strong historical information memory ability.The second is to propose Biogeography-based Optimization(BBO)algorithm to optimize the weight and bias of the network when considering its good migration mechanism and strong search capability.The third is to propose PSO and BBO fusion optimization algorithm to optimize the weight and bias of the network in order to make the final classification effect go a step further.Finally,this thesis verifies the feasibility and effectiveness of the proposed method on the same dataset.The comparison of PSO+BBO-BPNN with PSO-BPNN and BBO-BPNN for disease brain detection proves that the method in this thesis has relative superiority in MR brain image classification.In the experiment,a 10-fold cross-validation method aims at preventing overfitting.The average recognition rate of "WE+PSO+BBO-BPNN" for magnetic resonance brain images reaches 83.16%,and the average classification time is 43.68s.The experimental results show that the research method in this thesis can improve the efficiency of computer classification while ensuring the detection performance.