Research On The Application Of Tensor Network Machine Learning In Chest X-Ray Images

As a respiratory disease,pneumonia is attached more importance by countries around the world for its contagion and mortality.Early detection and treatment of pneumonia will be helpful to reduce preventable deaths.The observation and analysis of chest X-ray images by doctors is currently an effective diagnostic method for pneumonia.However,Experienced doctors take about 5 to 15 minutes of visual analysis of the patient’s chest X-ray images in past.It will bring huge pressure to the doctor when the hostipal is overcrowded with patients.Therefore,the clinical image diagnosis of artificial intelligence for pneumonia is an incoming trend.However,deep learning,as an important way to realize artificial intelligence,also introduces more complicated models while bringing high accuracy.These models not only depend on large-scale datasets,but also heavily rely on the computing resource of hardware.In order to aviod defects in deep learning represented by convolutional neural networks in diagnosing the X-ray images,this article proposes and studies the tensor network machine learning model.The role of tensor networks in diagnostic chest X-rays is illustrated from generation and discrimination tasks respectively,as follows:(1)Build a batch generative network classifier for diagnosis of pneumonia in chest X-ray image.The generative tensor network classifier has achieved excellent results on MNIST,but still not enough to process medical image in practice.Firstly,this chapter introduces a special histogram equalization algorithm to enhance the X-ray images.Secondly,we bring the locality in batch generative tensor network classifier,which improves the ability of feature extraction.The accuracy of 87% was obtained in the pneumonia detection experiment,not only exceeding the original model,but also far ahead of other classical machine learning algorithms.(2)Introduce the generative tensor network FT-Nets based on the tensor layer built on tensor mod product.In addition,the TNets architecture is proposed to process the multi-classification problems in image recognition.The calculation form of the tensor layer is similar to Tucker decomposition,and its parameters can be updated by gradient through back propagation.Compared with the weight of the neural network,the tensor layer can retain the original structural information of the input.The TNets built by tensor layer performed well in the experiment.The TNets has less parameters but strong ability to learn from data,compared with the full connection neural network of the same layer.TNets achieve the accuracy of 97.1% in X-ray image classification.(3)The generation and discrimination system is set up,integrated the above different algorithm models,and applied to the processing of X-ray chest images to achieve auxiliary diagnosis of pneumonia.The generation and detection system of tensor network is developed based on the Vue framework.When deploying the project,the page will be rendered according to the template file.The Flask framework is work as backend,which is not only responsible for the run of model,but also interacts with the frontend of the message and file.This system can not only configure a tensor network model online,but also verify the algorithm model online.Finally,the main functions of the system are tested to basically meet design expectations.This system can provide important advice for clinicians to diagnose pneumonia according to X-ray images.