The Intelligent Diagnosis Method of Covid-19 Based on the Lenet-Vit Deep Neural Network

Abstract

The method for intelligent diagnosis of COVID-19 based on the LeNet-ViT deep neural network was proposed. The LeNet-ViT model was created, it has the following advantages: the input image is not square, which expands the scope; the input image is pre-compressed and the new size depends on the original image size, and it is empirically determined, which increases the model training speed and the model identification accuracy; the number of pairs “convolutional layer - downsampling layer” depends on the image’s size, and it is automatically determined, which increases the model classification accuracy; the number of layer planes is automatically determined, which speeds up the definition of the model structure; the patch size depends on the image size, and it is empirically determined, which increases the model identification accuracy; the number of encoder blocks is empirically determined, which increases the model learning speed; the use of a convolutional neural network allows to efficiently extract features, and the use of a visual transformer allows to effectively analyze these features. The proposed method for intelligent diagnosis of COVID-19 can be used in various intelligent computer systems for medical diagnostics.