Study On The Application Of CT Radiomics And Whole Lung Gray Histogram Analysis Parameters In The Diagnosis And Treatment Of COVID-19

Part Ⅰ:Evaluation of clinical features and CT signs of COVID-19 and evaluation of different layersObjective1.To compare the differences of clinical characteristics and CT signs between the diagnosed COVID-19 patients and CT was positive patients in fever clinic,so as to improve the understanding and prevention of COVID-19.2.According to COVID-19 related CT signs,compare the influence of 5mm and 1mm slice thickness on early lesion identification in COVID-19.Materials and methodsA retrospective analysis of 1071 fever clinic and special CT examinations for special patients from January 24,2020 to March 31,2020.Finally,we included 61 confirmed patients in the COVID-19 group and 64 non-COVID-19 group patients with positive chest CT.The imaging information and clinical data of the patients were collected retrospectively through the PACS system and HIS system.The image information included CT range features,distribution features,morphological features,lesion density,lesion detail features and accompanying signs.Clinical data included gender,age,onset of symptoms,accompanying underlying disease,and laboratory examination.The differences between COVID-19 group and non-COVID-19 group were compared by Fisher’s exact test or chi-square test,and the differences between two groups of continuous variables were compared by Mann-Whitney U test or student t test.Two other radiologists who did not know the results recheck the chest CT images of two groups(61 in COVID-19 and 64 in non-COVID-19),evaluate the CT signs related to COVID-19 on 5mm and 1mm slices,classify the possibility of viral pneumonia in patients,and score the involved range of lesions.Kappa coefficient test and intra-group correlation coefficient test were used to test the consistency of two doctors.Mc Nemar test or Cochran Q test and paired sample rank sum test were used to compare the classification and continuous variables of 5mm and 1mm slices groups.Result1.In this study the median age of COVID-19 patients is about 55.The clinical manifestations are fever,cough,chest tightness,fatigue,cough with a small amount of white sputum,etc.The laboratory examinations are lymphocyte decrease,leukocyte decrease and C-reactive protein increase,etc.Compared with non-COVID-19 group patients with positive chest CT,the characteristic clinical symptoms were fatigue,muscle soreness,hypertension and decreased lymphocyte,leukocyte and hemoglobin in laboratory examination.2.Compared with the non-COVID-19 group,the common CT signs in the COVID-19 group were bilateral,multiple,involving multiple lobes,relatively large score of involved area,spherical ground glass density shadow,pure/predominantly ground glass density lesions,thickening of interlobular septa and mainly distributed in the periphery.The rare or absent CT signs include the lobar segment distribution,mainly consolidation,core lobular nodules,tree buds and acinus nodules,accompanied by bronchial wall thickening,pleural effusion and mediastinal lymph node enlargement.3.There is no significant difference between 5mm and 1mm layer thickness in the identification of COVID-19-related density features and detail features and the classification of the possibility of viral pneumonia,and it has high accuracy.The score of 1mm thickness on the extent of lesion involvement is higher than that of 5mm thickness.There is a high similarity between two doctors in the consistency test of layer thickness correlation.Conclusion1.COVID-19 has certain characteristics in early clinical symptoms and laboratory tests compared with suspected COVID-19 patients in fever clinic.The chest CT signs of COVID-19 can assist clinical diagnosis and screening.2.In order to improve the efficiency,the COVID-19 was screened on the thickness of 5mm according to CT signs during the epidemic,which could meet the preliminary diagnosis.However,there is a risk of missed diagnosis of lesions with a thickness of 5 mm,the evaluation of lesion range must be made on the thickness of 1 mm.Part Ⅱ:Study on differentiating COVID-19 from influenza A virus pneumonia based on plain CT radiomics modelObjective1.To compare the clinical data and chest CT findings of lung infection with COVID-19 and Influenza A.2.To explore the value of chest plain CT-based radiomics model in differentiating COVID-19 from influenza A virus pneumonia.Materials and methodsA total of 74 patients with pneumonia caused by influenza A virus from November 2018 and November 2019 to March of the following year were analyzed retrospectively.A total of 61 patients diagnosed in COVID-19 from January 24,2020 to March 31,2020 were analyzed retrospectively.The difference between two groups of categorical variables COVID-19 and influenza A virus was compared using Chi-square test or Fisher’s exact test,and the difference between two groups of continuous variables was compared using Student’s t test or Mann-Whitney U test.Two groups of patients’ chest CT plain images were analyzed,Influenza A virus pneumonia 64 and COVID-19 pneumonia 61,the lesions were manually delineated and the imaging characteristics were extracted.Patients were randomly divided into training group(46 Covid-19+48 Influenza A)and verification group(15 Covid-19+16 Influenza A).Saliency analysis,Peason correlation analysis and LASSO regression analysis were used to screen the key features.Three algorithms,namely,Logistic Regression(LR),Support Vector Machine(SVM)and Gaussian Naive Bayes(NBG),were used to build the diagnosis model in the training group and test it in the verification group.Plot the calibration curve and calculate the Hosmer-Lemeshow test to evaluate the fit of the model.Receiver operating characteristic curve(ROC)was used to analyze the diagnostic efficiency of the model,and Delong’s test was used to test the AUC significance of the three models.Decision Curve Analysis(DCA)was used to evaluate the clinical value of the model.Results1.Influenza A virus has a higher degree of fever,and chest pain symptoms are only seen in this group.The probability of lymphocyte decrease is higher than COVID-19,and the proportion of monocyte increase,eosinophil decrease and C-reactive protein increase is significantly higher than that of COVID-19.Influenza A virus has many overlapping signs with COVID-19,but some signs such as lobar distribution,predominantly consolidation,enlarged mediastinal lymph nodes,and pleural effusion only occur in patients with Influenza A.2.For the discrimination between COVID-19 and Influenza A pneumonia,11 imageomics were screened out by LASSO dimension reduction to construct the diagnostic COVID-19 model.The AUC,sensitivity and specificity of LR model are 0.903,92.73%and 82.26%in the training group and 0.888,91.30%and 74.07%in the verification group.The AUC,sensitivity and specificity of SVM model are 0.937,96.36%and 85.48%in the training group and 0.923,91.30%and 77.78%in the verification group.The AUC,sensitivity and specificity of NBG model are 0.929,91.82%and 80.65%in the training group and 0.889,97.83%and 70.37%in the verification group.The diagnostic efficiency of SVM model is slightly higher than the other two models.Calibration curve analysis shows that all three models have good fitting degree.DCA analysis proves that the net income of the three models is basically the same,and all of them have clinical value.3.The efficiency of SVM radiomics model in diagnosing COVID-19 is slightly higher than the other two models,and the difference is not statistically significant.Calibration curve analysis shows that the three models have good fitting degree,and DCA analysis proves that the net income of the three models is basically the same,and they all have clinical value.Conclusion1.The clinical manifestations of influenza A virus pneumonia are more virulent and inflammatory.It is challenging to distinguish the image of Influenza A from COVID-19,especially the severe and critical type,but the large lamellar consolidation with interstitial inflammation,mediastinal lymph node enlargement and pleural effusion are beneficial to distinguish.2.Plain CT-based radiomics model,the accuracy of distinguishing COVID-19 from Influenza A pneumonia is high,which has certain clinical value.Part Ⅲ:Application of parameters based on whole lung gray histogram analysis in clinical classification and early prognosis of COVID-19Objective1.To explore the factors related to clinical classification of COVID-19 patients’clinical data,CT semi-quantitative analysis and computer-aided whole lung gray histogram analysis quantitative parameters,and to build a nomogram model to identify clinical severe and critical patients.2.To compare the clinical data of COVID-19 patients with normal and abnormal chest CT images at the time of discharge from hospital,and the differences of quantitative parameters in CT semi-quantitative and computer-aided whole lung gray histogram analysis,so as to explore the indicators for predicting the normal chest CT image at the time of discharge from hospital in COVID-19.Materials and methodsRetrospective analysis was conducted on 61 patients diagnosed with COVID-19 on January 24,2020 and March 31,2020,including common type(29 cases)and severe and critical type(32 cases),the two groups of patients were randomly divided into a training set(20 cases of ordinary type,and 21 cases of severe and critical type)and a verification set(9 cases of ordinary type,and 11 cases of severe and critical type).Patients were divided into normal group(17 cases)and abnormal group(44 cases)according to CT images at discharge.Clinical data and CT images of patients were collected by HIS system and PACS system,and the lesions involved area was scored independently by 3 experienced radiologists using a double-blind method to obtain the semi-quantitative value of CT images.The proportion of lesion volume was obtained by using the pneumonia auxiliary diagnosis system authorized by ShuKun Technology Company(Version 1.20.0)and whole lung gray histogram analysis quantitative parameters was obtained by Pulmo 3D software of Siemens Healthineers.In the study of severity,all clinical data,semi-quantitative values and computer-aided quantitative parameters of CT were first subjected to univariate Logistic regression,and the relevant variables with P<0.05 were selected.After that,Multi-factor Logistic regression(stepwise regression method)was used to screen out important variables related to severity.R software was used to construct the normogram model for predicting severe and critical patients in the training set.Receiver operating characteristic curve(ROC)was used to evaluate the diagnostic efficiency of nomogram in training set and test set respectively.Draw the calibration curve in the verification set and evaluate the consistency of nomogram by Hosmer-Lemeshow goodness of fit test.The clinical value of nomograms was assessed in the validation set using the Decision Curve Analysis(DCA).In the CT images at discharge study,univariate Logistic regression was performed on clinical data of all patients,semi-quantitative values of CT images and computer-aided quantitative parameters of CT,and ROC analysis was used to analyze the diagnostic efficacy of relevant single factors,and the threshold was selected according to the maximum Jordan index.Chi-square test or Fisher’s exact test,Mann-Whitney U test were used to evaluate the differences in clinical data,CT semi-quantitative and quantitative values between ordinary type,severe and critical type and patients with normal and abnormal chest CT at discharge.Result1.The results of univariate Logistic regression analysis related to severity showed that the age of clinical data,whether there was hypertension or not,the number of white blood cells,lymphocytes,neutrophils,hemoglobin,D-dimers,prothrombin time,CRP,LDH levels,and all CT semi-quantitative and computer-aided whole lung gray histogram analysis quantitative parameters were related to clinical classification(P<0.05).Multi-factor Logistic regression finally involved three variables,namely area score,Standard Deviation(SD)and Low attenuation value percent(LAV%)of whole lung gray histogram analysis quantitative parameters,into the regression equation(P<0.05),and the nomogram model of predicting severe and critical type was constructed in the training set.ROC curve analysis shows that the area of nomogram under the curve of training set and test set is 0.974(95%CI,0.869-0.999)and 0.929(95%CI,0.772-0.995),respectively,which has good diagnostic efficiency.The positive curve drawn on the test set shows that the predicted probability has a good fit with the actual probability.DCA analysis in test set proves that nomogram model has clinical value.2.For patients with normal and abnormal chest CT at the time of discharge,there were significant differences in clinical factors such as age,laboratory lymphocyte and neutrophil counts at the time of admission,levels of hemoglobin,D-dimer,C-reactive protein and lactate dehydrogenase,as well as semi-quantitative CT imaging and computer-aided quantitative CT parameters at the time of admission(P<0.05).The results of univariate Logistic regression and ROC analysis showed that the Mean lung density(MLD),a quantitative parameter in the whole lung gray histogram analysis,was the best for predicting the diagnostic efficiency of patients with normal lung tissue at discharge,and the threshold was-782HU.Conclusion1.Compared with clinical data and laboratory tests,the quantitative parameters analyzed by chest CT whole lung gray histogram at the time of admission have a stronger ability to predict the clinical severity of COVID-19 and lung tissue at the time of discharge.2.According to the CT semi-quantitative and computer-aided quantitative parameters at the time of admission,it is effective to construct a nomogram model for predicting the severe and critical COVID-19.CT computer-aided quantitative parameter MLD at the time of admission could predict the situation of patients in COVID-19 at the time of discharge.