Research On Automatic Recognition And Grading Of Mucus Impaction In CT Images Based On Deep Learning

Along with the continuous increase in the number of asthma patients,the amount of related medical imaging data is rapidly increasing,which results in increasing pressure for manual diagnosis in clinical practice.With its high-efficiency,objective and powerful advantages,artificial intelligence has become the key method to solve the problem.As more and more progress in intelligent medicine,the purpose of this paper is to utilize deep learning techniques to analyze CT images and aid doctors in diagnosing and treating patients.It specifically concentrates on two important research objectives:automatic recognition and intelligent grading of mucus impaction.The automatic recognition task of mucus impaction belongs to the problem of irregular shape lesion detection,and it focus on improving the model and solving the problem.Due to the irregular shape,uneven gray scale and small airway distribution of the mucus impaction,the accuracy of human eye recognition is low.In this paper,we design and implement an automatic recognition algorithm of mucus impaction based on the improved Faster R-CNN,which does great help to locating the lesion.Firstly,deformable convolution is added to the backbone network to process global features,and effective information is extracted by adaptive receptive field.Secondly,feature pyramid network with weight coefficient is used for fusing multi-scale features to fill the semantic gap between different feature layers.Finally,the feature map is normalized by deformable ROI Pooling,and the feature shift in deformable convolution is automatically learned to assist in the detection.Conduct experiment designs to validate that this algorithm superior detection outcomes in comparison to other existing models.The proposed automatic recognition algorithm for mucus impaction can relieve doctors’ burden of CT images reading,improve diagnostic efficiency,and provide a basis for the development of subsequent treatment plans.The intelligent grading of mucus impaction belongs to the problem of auxiliary disease diagnosis,and it focus on analyzing and solving complex problems reasonably and effectively.At present,it is inefficient and subjective to use manual scoring method to get the grading results of mucus impaction.In this paper,we design and implement a fully automatic endto-end intelligent grading of mucus impaction auxiliary diagnosis scheme,which is specifically divided into three parts:automatic recognition of mucus impaction,intelligent segmentation of lung lobes and intelligent grading of mucus impaction.Firstly,the above automatic recognition algorithm is used to detect the mucus impaction;Secondly,lung lobe segmentation is implemented based on Swin-Unet model;Finally,combined with the results of mucus impaction detection and lung lobe segmentation,the intelligent grading results of mucus impaction are obtained based on the clustering algorithm.Through Spearman correlation analysis,it can be concluded that the intelligent grading scheme of mucus impaction in this paper is reasonable and reliable,providing auxiliary reference for patients’ disease evaluation.