Research On Lung CT Lesion Segmentation Based On Deep Learning

In the early stage of lung cancer,it is not easy to detect the lesion area through CT,leading to many patients missing the optimal treatment opportunity.However,the initial CT obtained due to dim ambient light results in low brightness and cannot provide reliable information.Traditional Convolutional neural network performs well in medical focus segmentation,but the extracted feature information cannot further improve the segmentation accuracy.The semantic differences between the lesion area and normal tissues are low,making it difficult to segment pneumonia infected areas.Deep learning can provide richer feature information for image segmentation,but there is limited research on its application in lung CT lesion segmentation.In view of this,this paper conducts research on lung CT lesion segmentation based on deep learning,and intends to use improved deep learning technology to solve the above problems.The specific innovative points and main work are as follows:1.For the problems of insufficient brightness and loss of feature information in the initial lung CT,an R-Retinexnet image enhancement method is proposed,which converts the initial image from RGB domain to HSV color space,enhances the color of the V component illumination image,and finally returns to RGB domain.After processing,the PSNR value of the image was increased to 30.41 and the gradient value was increased to 7.795,indicating that R-Retinexnet achieved higher image quality than MSRCR,BSSR,and V-SSR methods.2.In order to obtain a better segmentation accuracy,a Goog Le Net Resnet network that combines two types of deep learning networks,Goog Le Net and Resnet,and can be convolved on multiple scales is proposed.Through the superposition of feature information,the network Rate of convergence is improved.The segmentation accuracy of pneumonia CT reaches 95%,which has a better segmentation effect than Alexnet,Resnet and other networks,and can accurately distinguish three types of people: normal,ordinary pneumonia,and severe pneumonia.3.To address the difficulty in accurately identifying the lesion edges of pneumonia CT,an EA-Unet segmentation method is proposed,which introduces attention mechanism during the skip connection process based on traditional Efficient Nets and U-net networks.After verification,this method achieves higher PA and MIOU values when processing pneumonia CT compared to the U-net segmentation method,significantly improving segmentation accuracy and providing more accurate localization for pneumonia diagnosis.