Key Techniques Research Of Multiscale Brain Images Based On Machine Vision

Computer-aided medical diagnosis combines medical diagnosis and artificial intelligence.It can be applied in clinical diagnosis,imaging-based diagnosis,etc.Its ultimate aim is to increase the diagnosis accuracy.As the developing of modern medicine,each patient will provide a massive of different types of information during diagnosis,especially the imaging data.How to fully explore the imaging data is an important problem.This research of machine vision-based multlscale brain image processing may provide a possible solution to this problem.This dissertation focuses on automatic diagnosis methods of multiscale brain image.Nowadays,hospitals tend to acquire millimeter-scale image over region of interest,and then the radiologists make decision over these images.Nevertheless,some diseases influence neurons other than brain structure.Hence,we need higher-resolution imaging technique than commonly used.This dissertation employed magnetic resonance imaging(MRI)to acuiqre millimeter-scale images,and confocal laser scanning microscopy to acquire micrometer-scale images.Current diagnosis methods only work for millimeter-scale images.Their extracted features are not robust,perform not well in classification,and only apply in specific disease.Our contribution is three folds:(i)We developed smart diagnosis systems oriented for multiple brain diseases;(ii)We proposed novel image features for brain slices;(iii)We constructed new algorithms for micrometer-scale images.The main highlights are listed below:? Traditional brain diagnosis systems only apply in specific disease.including glioma,meningioma,Alzheimer's disease,Alzheimer's disease plus visual agnosia,Pick's disease,sarcoma,Huntington's disease,chronic subdural hematoma,cerebral toxoplasmosis,herpes encephalitis,and multiple sclerosis.We proposed to use stationary wavelet transform to extract features of MR brain image,and validated its effectiveness.We proposed three hybridization methods of artificial bee colony and particle swarm optimization,in order to train the classifier.The simulation experiment showed our method is superior to thirteen state-of-the-art methods.?The sensorineural hearing loss is a serious brain disease.The alternation of brain structure is slight and unperceivable.At present,there are no diagnosis systems for hearing loss.This dissertation proposed a novel algorithm.It used fractional Fourier transform to extract brain features,and transform the image to unified spatiofrequency domain.We used principal component analysis for dimensionality reduction.Finally,we used the single-hidden-layer feedforward neural network as the classifier.The dataset contains the brain imaging data of 29 hearing loss patients and 20 healthy controls.The results showed our method is effective,and its performance is better than state-of-the-art methods.?The traditional millimeter-scale images cannot identify micrometer-scale change.There are many immuature spine within the brains of intellectural disability children;hence,we need to check in micrometer-scale images.We proposed a novel neuron-diagnosis system,based on micrometer-scale neuron images,and it can identify types of spines.Our experiment used the mouse neuron as the samples.We proposed two different algorithms to backbone extraction,edge location,and spine classification(mushroom,stubby,thin).Experiments showed the effectivess of our two algorithms.In all,this dissertation employed machine learning,oritened for brain and neuron structural images,implemented researches for computer-aided diagnosis techniques of mutilscale brain images.Besides,we focused on the smart diagnosis algorithms.Through the multiscale-related researches,we not only detected the diseases altering brain strcutures in millimeter-scale,but also diseases altering neurons in micrometer-scale.The machine vision methods help us obtain higher diagnosis accuracy than traditional methods.The combination of multiscale and machine vision can shed light on new smart brain-disease detection methods.Our method can help us improve prevention and diagnosis works of brain diseases.Furthermore,it can provide diagnosis standards for medical diagnosis and analysis.It can support medical data for medical statistical work.It may find physiological meaning for brain diseases,and accelerate related research works in biomedical fields.The researches can help accelerate the developments of next-generation artificial intelligence technique,brain computing,and information fields.