An Auxiliary Diagnosis System For Brain Diseases Based On The Uneven Distribution Of Medical Images

Brain tumor is one of the most common major causes for the increase in Mortality among human.At present,the diagnosis of brain-based medical images mainly depends on the doctor's visual interpretation.When the doctor performs a naked eye analysis,it is not easy to find many tiny texture changes and morphological features in the brain medical image,which will affect the early judgment of the condition.Therefore,doctors need to use brain disease auxiliary diagnosis system to improve the accuracy of brain tumor diagnosis and reduce the rate of missed diagnosis.The traditional classification algorithm assumes that the data set of the training sample is balanced and the misclassification costs are equal.Usually,the error rate is minimized.However,brain medical image diagnosis has the characteristics of class unbalance and unequal misclassification costs.Therefore,when the traditional classification algorithm uses a clinically diagnosed MRI brain medical image as a training set to construct a classification model,the classification effect is poor and easy to be insensitive to positive.It is difficult for the auxiliary diagnosis system for brain diseases to have high accuracy and weak generalization ability.In order to improve the performance of the brain disease auxiliary diagnosis system,this paper introduces costsensitive mechanisms,A Cost-sensitive Probabilistic Neural Network is designed based on a traditional cost-insensitive probabilistic neural network based on density function kernel estimation.So the problem of class unbalance and unequal misclassification costs solved in MRI brain medical images.In order to develop an auxiliary diagnostic system for brain diseases with enhanced generalization ability,so to improve the accuracy of brain tumor diagnosis and reduce the rate of missed diagnosis.The paper develops a brain disease-assisted diagnosis system based on unequal category distribution of medical images.Because the median filter algorithm removes noise from the MRI brain image,it does not affect the image edges.First,median filtering algorithm is used to denoise the MRI brain medical images obtained from DICOM 3.0 interface.Then use the pulse coupled neural network to segment the denoised MRI brain medical image.Segmented images are more conducive to accurate and effective classification.Then using discrete wavelet transform to extract the wavelet coefficients from the MRI brain images segmented by the pulse-coupled neural network as feature vectors,which will be used as the input of the classification model and the basis for constructing the classification model.Since the extracted wavelet coefficients are related to each other and too many features increase the storage space and calculation time,principal component analysis is used to reduce the dimension of the wavelet coefficients extracted by the discrete wavelet transform to obtain lowdimensional feature vectors.Finally,the low-dimensional feature vector with class unbalance distribution is used as the training set.The classification model is constructed for the training set using the cost-sensitive probability neural network proposed in this paper,and the unknown classification is set using the constructed classification model.MRI brain medical images classified as either normal or abnormal.Using 120 MRI brain medical images as experimental data to assess the performance of the brain disease assistant diagnostic system developed in the paper,the experimental results show that the brain disease auxiliary diagnosis system developed in this paper not only has higher accuracy and stronger generalization ability,but also is very sensitive to high-cost positive categories.