Pathological Brain Detection By Extreme Learning Machine Optimized By Bat Algorithm

Magnetic resonance imaging(MRI)is widely used for clinical check.MRI can provide high resolution images on soft tissues,so it is often applied in brain imaging.MRI plays an important role in early detection and treatment of brain diseases.As a 3D imaging modality,a brain MRI contains massive information which requires professional knowledge to interpret.However,manual interpretation is time consuming and tedious,and mistakes happen inevitably.Automatic pathological brain detection(PBD)has attracted more and more attention with the development of computer vision and artificial intelligence.PBD systems can classify the input brain MRIs as healthy or pathological automatically and accurately,and even recognize what exact disease the brains have,which is helpful for the doctors and physicians.PBD systems can increase the diagnosis accuracy of doctors and help relief the pain of patients.Over the last decade,scholars have devoted their efforts in this field,and made improvement.In this thesis,I proposed a novel pathological brain detection system based on brain MRIs.Firstly,I performed 2 level 2D discrete wavelet transform on the images.Then,I calculated 7 Shannon entropies from the obtained wavelet sub-bands to form the feature vector of each image,namely wavelet entropy(WE).Single hidden layer feedforward neural network is selected as the classifier,which was trained by extreme learning machine(ELM).Classical training algorithms of networks usually work with gradient based descent and iterations,but ELM works with only random hidden nodes and generalized inverse matrix.Therefore,ELM converges much faster and the generalization ability is promising.Finally,I employed bat algorithm(BA)to optimize the parameters of the ELM,and further improved its generalization ability and robustness.Totally 132 brain MRI samples obtained from the Website of Harvard University Medical School was employed to evaluate the performance of my method.In experiment,10×10-fold cross validation was used.Experimental results suggested that the proposed WE+BA-ELM achieved 98.33%,99.04%,and 93.89%for accuracy,sensitivity and specificity,respectively.I also compared WE+BA-ELM with five state-of-the-art approaches,including DWT+PCA+RBFNN,WE+RBFNN,WE+KELM,WE+OS-ELM and FRFE+MLP+ST-Jaya,and found out that WE+BA-ELM performed the best among these methods in terms of classification ability.In this thesis,a pathological brain detection method called WE+BA-ELM was proposed.The detection performance of this system is state-of-the-art.WE efficiently reduces the feature space from the original 256x256(image size)to 7,which reduced both memory occupation and computation complexity.BA belongs to a new swarm intelligent algorithm,which can effectively search the solution space and find the global optimal solution.BA helped to improve the robustness of ELM.For future work,I shall add deep learning algorithms in,to detection what the specific disease the pathological brains are infected with and try to locate the focus of infection.