| COVID-19 is a respiratory disease with a high infection rate.Since its appearance,it has spread rapidly around the world.The number of people infected by the virus has increased dramatically worldwide,which has aroused great public health concern in the international community.At this time,the detection,prevention and control of the new coronavirus is extremely important.Computerized tomography(CT)imaging technology plays a vital role in the fight against global COVID-19.The performance of CT images can identify COVID-19 early,accurately and in a timely manner,which will improve the efficiency of treatment of COVID-19 cases.Preventing the spread of the epidemic and reducing the pressure on front-line medical staff and medical institutions are of great significance.The research contribution and innovation of this thesis is the use of deep learning semantic segmentation to detect the lesions in the chest CT images of COVID-19 patients in clinical medicine.In view of the limitations of existing deep learning methods in the segmentation of chest CT images of COVID-19 patients:first,the sample size of medical images is insufficient,and it is difficult to effectively train the network;second,the COVID-19 lesion has no fixed shape and no clear boundaries,the image segmentation effect obtained by using the existing neural network model is not good.This article proposes a CA-Unet segmentation method suitable for chest CT images of COVID-19 patients based on the network,which is mainly improved from three aspects:(1)Reduce the number of model channels to reduce the scale of the model,avoid over-fitting,and reduce the complexity of model calculation;(2)Add a multi-level feature fusion module to fuse high-level features with low-level features,Merge into more discriminative features,while solving the problem of insufficient sample data,improving the segmentation performance of the model;(3)Add attention mechanism to adjust the weight corresponding to each element of the feature map,thereby suppressing noise and useless information,and improving Image segmentation accuracy.This article uses clinical medical patients’ chest CT images to verify the CA-Unet segmentation method.Through comparative analysis with existing segmentation models,it proves that CA-Unet has better segmentation performance and can quickly and accurately segment patients’ chest CT images to achieve lesion detection.And by calculating the area and proportion of the lesion area and the lung area,the patients’ degree of disease can be intuitively reflected in the form of numerical values to achieve the purpose of assisting doctors in diagnosis. |
