COVID-19 Lung Related Clinical And Artificial Intelligence Research Based On CT

Objective:Since the outbreak of COVID-19 disease 2019(Coronavirus disease 2019),the rapid spread of the world,high transmission and mortality rate has caused the infection of people in many countries and regions,death cases are increasing.The COVID-19 outbreak has increased the need for medical resources and nucleic acid screening.As a supplementary examination,computed tomography(CT),the advantages of rapidity and accuracy in judging the type,degree and change of lung lesions,and plays an important role in the global fight against COVID-19.However,there are still several major problems in CT diagnosis and assessment of COVID-19:(1)The prognosis and mortality of patients of different ages are significantly different,but the influence of age on the occurrence and progression of COVID-19 is unknown,which leads to the insufficient interpretation of CT as a biological marker for the follow-up of COVID-19.(2)The rapid development of COVID-19 is characterized by large changes in lung lesions,which cannot be accurately quantified by the naked eye,resulting in inaccurate dynamic monitoring of the disease.In the midst of the COVID-19 outbreak,it is necessary to find a reliable and time-saving automatic detection and evaluation method for pneumonia,which is conducive to accurate assessment of the disease progression and change of COVID-19.(3)In clinical practice,different types of COVID-19will lead to different prognosis and treatment.However,in the diagnosis and treatment process,evaluation of severe cases requires accurate and quantitative imaging support,but at present,visual observation can only give a subjective impression,which brings difficulties to subjective and accurate clinical classification.Artificial Intelligence(AI)is used to accurately segment lesions and achieve accurate clinical classification.It can help severe patients to be classified early and get correct diagnosis and treatment as soon as possible.(4)In the CT image interpretation of COVID-19 patients,we found that it is difficult to distinguish COVID-19 from other pulmonary infectious diseases,which usually requires etiological examination.How to quickly screen and quantify CT features of different pneumonia lesions and accurately classify diseases is a key problem to be solved urgently.In addition,the existing image artificial intelligence systems only stay in the diagnosis of pneumonia stage,classified image features related to the diagnosis is still not to explain,therefore,to carry on the study for feature selection the interpretability of study is very necessary,in order to interpret the weighted image features related to classification results,and really apply AI to clinical practice.The purpose of this study was to:(1)search for the regularity of clinical or CT features of COVID-19 patients of different ages,so as to clarify the role of age in disease severity and disease progression of COVID-19 patients.(2)establish an AI system for automatic detection and assessment of pneumonia.And test the accuracy of automatic AI assessment of COVID-19 lung involvement and find a more accurate method than artificial vision assessment,which will help the radiologists quickly and accurately assess the severity of COVID-19.(3)based on the multi-center data set,the established pneumonia automatic detection and evaluation AI system was used to explore the chest CT imaging characteristics and further quantitative and accurate evaluation of different clinical types of COVID-19 were discussed,in order to help early identification of severe risk factors.(4)establish an AI system based on deep learning(DL)method to classify infection types through CT images of pneumonia patients,as well as to present an effective clinically relevant machine learning(ML)system based on medical image identification and clinical feature interpretation to assist clinicians in triage and diagnosis.Materials and Methods:1.Comparative observation of clinical and CT signs of COVID-19 in different age groups:The 185 patients was confirmed COVID-19 were divided into 3 groups according to age:(≤30 years(28 cases),31～50 years(90 cases)and>50 years(67 cases)).Clinical data and plain chest CT images of all patients were retrospectively analyzed.χ~2or Fisher exact probability method and Kruskal-Wallis rank sum test,Mann-Whitney test were used to compare the clinical and CT characteristics of patients in different age groups.2.Establishment and effectiveness evaluation of an artificial intelligence system for automated COVID-19 detection and scoring:In this multicenter case series,chest CT plain scan images of 199 patients with confirmed COVID-19 from China were retrospectively analyzed in two methods:artificial vision assessment and automatic AI pneumonia assessment system.We used the area under Receiver operating characteristic(ROC)curve(AUC)to compare the diagnostic effectiveness of the two in diagnosing severe and critical COVID-19.Then,the combination of AI with artificial drawing assessment was proposed as a test method(called the comprehensive combination of criteria).The intraclass correlation coefficient(ICC)with the comprehensive combination of criteria assessment was used to assess the reliability of artificial vision assessment and computer automatic assessment,and the Bland-Altman plots with 95%limits of agreement were used to assess their agreement.3.Imaging characteristics analysis of different clinical types of COVID-19 based on AI:A total of 164 patients confirmed COVID-19 were retrospectively enrolled from multiple hospitals.All patients were divided into the mild type(136 cases)and the severe type(28cases)according to their clinical manifestations.The total CT severity score and quantitative CT features were calculated by AI pneumonia detection and evaluation system with correction by radiologists.The clinical and CT imaging features of different types were analyzed.4.Clinical application of AI system based on deep learning algorithm to identify pulmonary infectious diseases:The 3463 CT images of pneumonia used in this multi-center retrospective study were divided into four categories:bacterial pneumonia(n=507),fungal pneumonia(n=126),common viral pneumonia(n=777),and COVID-19(n=2053).We used DL methods based on images to distinguish pulmonary infections.A ML model for risk interpretation was developed using key imaging(learned from the DL methods)and clinical features.The algorithms were evaluated using AUCs.Results:1.Comparative observation of clinical and CT signs of COVID-19 in different age groups:1)C-reactive protein(CRP)(39.94±55.10 mg/L)increased in this group of COVID-19patients,while CRP(P<0.001 P<0.001)and lymphocyte count in>50 years old group were statistically significant compared with other groups(P=0.002,P=0.007).2)The clinical severity was significantly different among the three groups(P<0.001),and the>50 years old group included larger distribution of severe(69.0%)and fatal(100%)types than other groups.3)On the CT findings,the group≤30 years old had more ground lesions and less fan-shaped lesions,crazy paving pattern and other changes like emphysema,while the>50years old group found the opposite.4)There were significant differences between the>50 and the≤30 years old groups in all lung scores(P=0.001,right upper lobe;P=0.001,right middle lobe;P=0.025,right lower lobe;P=0.002,left upper lobe;P=0.031,left lower lobe;P=0.001,total severity score).2.Establishment and effectiveness evaluation of an artificial intelligence system for automated COVID-19 detection and scoring:1)The AI was superior to three radiologists in the diagnosis of severe and critical COVID-19,among which AI had an AUC of a total lung severity score for diagnosing severe-critical type was 0.898(95%CI 0.826-0.971),and the AUC of senior radiologists was0.879(95%CI 0.804-0.954),which was better than the two junior radiologists(0.871(95%CI0.795-0.947)and 0.849(95%CI 0.768-0.930)).2)For artificial vision assessment,the ICCs with the comprehensive combination of criteria assessment of two junior radiologists(0.8192(95%confidence interval[CI]:0.7675,0.8603))and 0.8765(95%CI:0.8397,0.9053)respectively)and senior radiologist(0.9087(95%CI:0.8810,0.9302))were lower than AI pneumonia assessment(0.9177(95%CI:0.8925,0.9372)).3)Bland-Altman plots showed that the AI pneumonia assessment had a narrowest 95%limits of agreement(-2.6072,2.1504)and the number of points out of limits of agreement for AI pneumonia assessment was less than artificial vision assessment by senior radiologist.3.Imaging characteristics analysis of different clinical types of COVID-19 based on AI:1)It was observed that patients in the severe type group were older than the mild type group.2)Round lesions,fan-shaped lesions,crazy-paving pattern,fibrosis,"white lung",pleural thickening,pleural indentation,mediastinal lymphadenectasis were more common in the CT images of severe patients than in the mild ones.3)Using the established AI system for automatic detection and evaluation of pneumonia,a higher total lung severity score and scores of each lobe were observed in the severe group,with higher scores in bilateral lower lobes of both groups.Further analysis showed that the volume and number of pneumonia lesions and consolidation lesions in overall lung were higher in the severe group,and showed a wider distribution in the lower lobes of bilateral lung in both groups.4.Clinical application of AI system based on deep learning algorithm to identify pulmonary infectious diseases:1)the median AUC of DL models for differentiating pulmonary infection was 99.5%(COVID-19),98.6%(viral pneumonia),98.4%(bacterial pneumonia),99.1%(fungal pneumonia),respectively.By combining chest CT results and clinical symptoms,the ML model performed well,with an AUC of 99.7%for severe acute respiratory syndrome coronavirus 2(SARS-Co V-2),99.4%for common virus,98.9%for bacteria,and 99.6%for fungus.2)Regarding clinical features interpreting,the model revealed distinctive CT characteristics associated with specific pneumonia:in COVID-19,ground-glass opacity(GGO)(92.5%;odds ratio[OR],1.76;95%confidence interval[CI]:1.71-1.86);larger lesions in the right upper lung(75.0%;OR,1.12;95%CI:1.03-1.25)with viral pneumonia;older age(57.0 years±14.2,OR,1.84;95%CI:1.73-1.99)with bacterial pneumonia;and consolidation(95.8%,OR,1.29;95%CI:1.05-1.40)with fungal pneumonia.Conclusion:1.In patients with COVID-19,the age distribution could affect both the clinical and CT features in significant measure.Taking consideration of the age influence could facilitate the hierarchical diagnosis and treatment of COVID-19 patients.2.Interobserver agreement for automatic AI pneumonia assessment was higher than artificial vision assessment,that the automatic AI pneumonia assessment system and combination of AI with artificial drawing assessment could be helpful in severity assessment for COVID-19 quickly and accurately.3.Chest CT of patients with severe COVID-19 pneumonia showed more consolidative and progressive lesions.With the assistance of AI,CT could evaluate the clinical severity of COVID-19 pneumonia more precisely and help the early diagnosis and surveillance of the patients.4.For classifying common types of pneumonia and assessing the influential factors for triage,our AI system has shown promising results.Our ultimate goal is to assist clinicians in making quick and accurate diagnoses,resulting in the potential for early therapeutic intervention.