Aided Diagnosis Based On Multimodal Medical Image Fusion And Machine Learning

Medical image fusion technology is a very characteristic application of imaging technology in medicine.It integrates both anatomical and functional images and achieves a combination of single-modality medical image diagnosis and multi-modality medical images.Comprehensive imaging diagnosis.This article will discuss an image fusion and an improved algorithm for classifying fused images to achieve the purpose of auxiliary diagnosis.The auxiliary diagnosis system is improved in terms of running time and accuracy.First,a medical image fusion algorithm based on wavelet transform is introduced.The basic idea of image fusion algorithm based on wavelet transform theory is to extract the detail information of two images and then fuse them according to the corresponding fusion rules.The image after fusion by the wavelet transform can present the information concerning the characteristics of the fusion of the first two images.Therefore,the image quality of the image can be improved.At the same time,we apply convolutional neural network method in machine learning to auxiliary diagnosis.Neural network technology is a new type of intelligent information processing technology that is formed by simulating the principle of biological nervous system.It has been successfully applied to the medical image classification,disease prediction,prescription compatibility and other medical fields.This paper mainly discusses an auxiliary diagnosis method based on multimodal image fusion and convolutional neural network.Because simply using a convolutional neural network for auxiliary diagnosis,time efficiency or accuracy cannot achieve particularly satisfactory results.Therefore,this paper uses an improved algorithm instead of the original convolutional network to achieve the purpose of improving the efficiency and accuracy of the auxiliary diagnosis.We first consider solving the problem of slow running time.Our solution is to use wavelet transform to obtain the low-frequency information matrix of image information,and use the low-frequencyinformation matrix as the input of the convolutional neural network to replace the input of the original image pixel matrix.Experiments show that this improved algorithm can effectively reduce the running time under the premise of ensuring the accuracy.Then we began to consider improving the accuracy of the auxiliary diagnosis.Because the information reflected by a single medical image is indeed insufficient in some cases to determine if it is ill.So next we use a multimodal medical image fusion method to fuse medical images of different modalities,use new multimodal medical images as training and test samples,and then submit them to an improved convolutional network for classification.Experiments show that this classifier has high classification accuracy.In the following,this paper considers the use of wavelet transform and convolutional network to combine the number of wavelet transforms and the choice of different fusion rules for wavelet transform fusion to improve the algorithm.We select three fusion rules for comparison experiments,which are the local variance fusion rule,the weighted average fusion rule,and the gray value maximum fusion rule for comparison experiments.The results show that the local variance fusion rule can indeed preserve the image information in the fusion process,thus providing a better information matrix to the convolutional neural network to ensure its accuracy.Next we discuss the effect of the number of wavelet transform layers on the experimental results through experiments.Taking into account the size of the sample image,we use one-level transformation and two-level transformation.The results show that based on the size of the image used in the experiment,the high-frequency information matrix obtained after one layer of wavelet transform is more suitable than the high-frequency information matrix obtained after two-layer transformation.According to the analysis,the high-level information matrices with too many layers may cause the loss of information.Although the input matrix size of the convolutional neural network can be reduced,the image information is sacrificed too much,so it is not recommended to pursue higher layers in the improved algorithm.Wavelet transform low frequency matrix.