Chest CT Evaluation For SARS-CoV-2 Infection And Development Of An Intelligent Diagnosis Model For Pneumonia Caused By Various Pathogens

Part Ⅰ Clinical and chest CT findings of patients infected with various variants of SARS-CoV-2Objectives Since January 2020,SARS-CoV-2 has undergone several mutations,and China has experienced pandemics caused by three main strains-the original strain,Delta,and Omicron.The Omicron strain has become dominant in the latest wave of outbreak peak.It is crucial to understand the pathogenesis and imaging manifestations of different SARSCoV-2 variants promptly for clinical diagnosis and treatment.Therefore,this study aims to compare the clinical characteristics and chest CT findings of patients infected with two important variants of concern and the original strain.Moreover,the severity of infection and lung involvement in patients with different vaccination statuses were also investigated.Methods Consecutive patients admitted to Wuhan Jinyintan Hospital,an infectious disease control hospital,were retrospectively analyzed at three different time points: the original strain from February 1st to 28 th,2020;Delta variant from August 1st to the 31 st,2021;and Omicron variant from December 7st to 13 th,2022.A total of 503 patients infected with the original variant(245 cases),Delta variant(90 cases),and Omicron variant(168 cases)were retrospectively analyzed.General demographics,clinical symptoms,clinical severity,laboratory examinations,and chest CT imaging findings were compared among these groups.The infection severity of patients with different vaccination statuses was also compared.Fisher’s exact test,ANOVA analysis of variance,or Mann-Whitney U test were used to determine differences between groups.Binary or multiple logistics regression was used to adjust confounding factors affecting group comparisons,including age,sex,underlying diseases,and vaccination status.Results The rate of severe disease significantly decreased from the original variant to the Delta variant and Omicron variant(original variant vs.Delta vs.Omicron: 27% vs.10%vs.4.8%,P<0.001).After controlling for confounding factors with multiple logistics regression,strain type was significantly negatively correlated with clinical severity(OR=0.306,P < 0.001).The rate of chest CT lesion detection was also decreased in the three groups [original variant vs.Delta vs.Omicron: 96%(235/245)vs.81%(73/90)vs.43%(73/168),P<0.001].A trend of gradual decrease in total CT score was observed across the groups [original variant vs.Delta vs.Omicron: 14(IQR 9.0-20.0)vs.6.0(IQR 3.0,8.5)vs.5.0(IQR 3.0-10),P<0.001].In the binary logistics regression analysis controlling for confounding factors,the total CT score>5 was significantly negatively correlated with strain type(OR=0.384,P < 0.001).Patients in the Delta variant group and Omicron variant group presented with "Indeterminate COVID-19 pneumonia" on CT compared to the original strain(original variant vs.Delta vs.Omicron: 1.3% vs.Delta,11% vs.Omicron,18%,P<0.001),And atypical COVID-19 pneumonia was significantly higher(original variant vs.Delta vs.Omicron: 1.7% vs.14% vs.16%,P < 0.001).The infection of the three strains mainly involved the lower lobes of both lungs(original variant vs.Delta vs.Omicron: 90% vs.78% vs.77%,P=0.007)and presented subpleural distribution(original variant vs.Delta vs.Omicron: 96% vs.85% vs.78%,P < 0.001),and the lesion form was dominated by GGO(original variant vs.Delta vs.Omicron: 75% vs.62% vs.66%,P=0.093).Both Delta and Omicron are mainly presented as solitary or multifocal ground-glass opacities(GGOs)or consolidations.Besides,patients infected with variant groups tended toward a higher prevalence of nodules,tree in bud,and halo signs than the patients with original variant infection(P<0.05 for all).Vaccinated cases had slighter clinical severity(P=0.04)and shorter in-hospital time [31(IQR 24,41)days vs.26(IQR 19,33)days,P=0.009] than the unvaccinated cases.Among Omicron patients,those who received a booster vaccine had a lower clinical classification compared to those who did not receive the booster vaccine(P=0.015).After adjusting for confounding factors,this difference was statistically significant(OR=0.313,P=0.009).Furthermore,Omicron patients who received a booster vaccine had less lung involvement compared to those who did not receive the booster vaccine(36% vs.57%,P=0.009).After adjusting for confounders,this difference was also statistically significant(OR=0.458,P=0.020).Conclusion Compared to the original strain and Delta variant,the Omicron variant exhibited lower clinical severity,less lung involvement on CT scan,and a tendency towards atypical lesions with less involvement.COVID-19 vaccination,particularly booster vaccinations,can effectively reduce the incidence of severe illness and lung involvement in patients infected with the Delta and Omicron strains and shorten hospital stays.Part II Long-term HRCT follow-up and lung function analysis of patients recovering from severe COVID-19 pneumoniaPurpose: The long-term prognosis of patients who have recovered from coronavirus disease 2019(COVID-19)is still not well understood,particularly those who had severe disease.Severe COVID-19 patients commonly exhibit extensive lung injuries during the acute phase of the illness.Consequently,this study aims to analyze the chest HRCT manifestations and lung function results of patients who were discharged from the hospital after experiencing severe COVID-19 at six months,12 months,and two years follow-up to better understand the recovery of lung anatomy and respiratory function.Furthermore,the study explores the risk factors that contribute to pulmonary fibrosis in convalescent COVID-19 patients.Methods: This study prospectively collected data from 114 cases of severe COVID-19 patients admitted to Wuhan Yintintan Hospital(n=69)and Wuhan Union Hospital(n=45)between December 25 th,2019,and February 20 th,2020.The 6-month follow-up study was conducted from June 20 th to August 31 st,2020.Patients with residual lesions on CT at six months were continued to be followed up for 12 months(December 20 th,2020,to February 3rd,2021)and two years(November 16 th,2021,to January 10 th,2022)after symptom onset.Lung changes(opacification,consolidation,reticulation,and fibrotic-like changes)and CT extent scores(score per lobe,0–5;maximum score,25)were recorded.Participants were categorized into two groups on the basis of their 6-month follow-up CT scan: those with CT evidence of fibrotic-like changes(traction bronchiectasis,parenchymal bands,lung structure deformation,and/or honeycombing)(group 1)and those without CT evidence of fibrotic-like changes(group 2).Between-group differences were assessed with the Fisher exact test,two-sample t-test,or Mann-Whitney U test.Multiple logistic regression analyses were performed to identify the independent predictive factors of fibrotic-like changes.In addition,patients who exhibited residual lung lesions on their 6-month CT scans were further followed up with HRCT scans at 12-month and 2-year intervals.Results: At 6-month follow-up CT,evidence of fibrotic-like changes was observed in 40 of the 114 participants(35%)(group 1),whereas the remaining 74 participants(65%)showed either complete radiologic resolution(43 of 114,38%)or residual ground-glass opacification or interstitial thickening(31 of 114,27%)(group 2).Multivariable analysis identified age of greater than 50 years(odds ratio [OR]: 8.5;95% CI: 1.9,38;P = 0.01),heart rate greater than 100 beats per minute at admission(OR: 5.6;95% CI: 1.1,29;P =0.04),duration of hospital stay greater than or equal to 17 days(OR: 5.5;95% CI: 1.5,21;P = 0.01),acute respiratory distress syndrome(OR: 13;95% CI: 3.3,55;P <0.001),noninvasive mechanical ventilation(OR: 6.3;95% CI: 1.3,30;P = 0.02),and total CT score of 18 or more(OR: 4.2;95% CI: 1.2,14;P = 0.02)at initial CT as independent predictors for fibrotic-like changes in the lung at six months.At the 6-month follow-up,27 patients(26%)showed abnormal pulmonary diffusion function(DLCO<80% prediction),with a higher prevalence observed among patients with fibrosis(50% vs.13%,P<0.001).CT at 12 months and two years of follow-up revealed a progressive decline in the proportion of patients displaying non-fibrotic residual pulmonary abnormality(44% at 12 months and 21.9% at two years).However,evidence of fibrotic-like changes persisted(P=1.00).Conclusion: Six-month follow-up CT scans showed fibrotic-like changes in the lungs in more than one-third of severe COVID-19 survivors.These changes were associated with older age,acute respiratory distress syndrome,longer hospital stays,tachycardia,noninvasive mechanical ventilation,and higher initial chest CT score.Those fibrosis-like changes persisted at 12-month,and two years followed up after discharge and were associated with reduced pulmonary diffusion function.Part III:AI-assisted HRCT was used to develop a diagnosis model for pneumonia caused by various pathogensPurpose: Identifying pneumonia pathogens in a clinical setting can be complex and challenging.Accurately identifying the causative pathogen is of significant importance for the diagnosis and treatment of pulmonary infectious diseases.Consequently,this study aims to develop an intelligent diagnosis system based on HRCT for identifying pneumonia pathogens.Method: This prospective study collected chest CT examination data from 2353 patients diagnosed with different pathogenic bacterial pneumonia at Wuhan Union Medical College Hospital between January 1st,2015 and December 31 st,2020.The patients were classified into five pneumonia categories and 12 subcategories based on a positive PCR test or sputum culture.The cases were divided into training and test sets with a 1:1 ratio(1166 vs.1187).External validation was conducted using 142 cases of viral pneumonia,50 cases of bacterial pneumonia,and 50 cases of fungal pneumonia collected from Wuhan Tongji Hospital and Wuhan Jinyintan Hospital.To classify the pathogen of a patient based on CT volume,we proposed the Deep Diagnostic Agent Forest(DDAF)algorithm.This is a challenging multiclass classification problem with large intraclass variations,small interclass variations,and imbalanced data.Notably,the level-I classification includes five categories of pneumonia: viral,bacterial,fungal,chlamydia,and mycoplasma infections.The level-II classification includes specific pathogens that cause pneumonia,such as SARS-CoV-2,cytomegalovirus,respiratory syncytial virus,Streptococcus pneumonia,Bacillus Corella,A.aureus,monocellular Pseudomonas aeruginosa,Acinetobacter baumannii,Aspergillus,Candida,mycoplasma and chlamydia.Additionally,seven radiologists with varying levels of experience participated in reader studies involving independent reading and AI-assisted reading.Based on their years of experience,the physicians were divided into three groups: the low seniority group(with an average seniority of four years),the middle seniority group(with an average seniority of 12 years),and the high seniority group(with an average seniority of 32 years).Results: The multi-way area under the receiver curves(AUC)for level-I diagnosis was 0.899±0.004,with an accuracy of 71.3±0.9%.For level-II differentiation,the multi-way AUC was 0.851±0.004,with an accuracy of 52.4±0.3%.Regarding external validation data for level-I authentication,the overall AUC was 0.7755,with an accuracy of 63.4%.The model outperformed the average results of the seven human readers in level-I recognition and also outperformed all readers in level-II recognition.Additionally,AI-assisted reading improved level-I classification by 6.8%,secondary classification by 4.3%,and viral classification by 6.7%.Conclusions: The accuracy of level-I and level-II diagnosis of pneumonia caused by different pathogens was higher with the deep forest network-based approach compared to manual diagnosis.AI-assisted manual diagnosis can also improve the efficiency of independent manual diagnosis.The deep learning model based on CT imaging scans accurately identifies pneumonia pathogens,providing a high-precision auxiliary diagnostic tool for pneumonia.