| Objective:1.Based on quantitative computed tomography(QCT),explore imaging quantitative indicators related to the prognosis of IPF and test the correlation between imaging quantitative indicators and pulmonary function parameters,providing new indicators for evaluating the prognosis of IPF.2.To explore radiomic features that can be utilized for predicting IPF patients’ prognosis based on CT radiomics,and develop,test and examine the radiomics predicting model,and examine the efficacy.Methods:1.IPF cases diagnosed at Henan People’s Hospital from January 2018 to December 2020 were collected,and categorized into a group of favorable prognosis or of unfavorable prognosis in accordance with inclusion、exclusion and grouping criteria.High-resolution CT images,pulmonary function and general clinical information,including sex,age,smoking history,smoking index,history of alcohol intake,BMI,and history of occupational exposure,were collected.All CT images were downloaded via the Hospital’s medical image system,and saved as DICOM.The composite physiologic index(CPI)were computed,including CPI1,which has been used frequently in the past,and CPI2,which has been revised by computer software.Medical image information system(Version:FACT-Lung 1.7.6.5)was utilized to delineate the region of interest(ROI),including honeycombing,reticular pattern,ground-glass shadow and bronchiectasis shadow.Honeycomb shadow and grid shadow are defined as pulmonary fibrosis areas.SPSS26.0 statistical software was adopted to conduct a contrast analysis of the differences of the clinical information of the two groups.2.Quantitative computed tomography:A quantitative calculation was conducted by means of medical image information system for the entire lungs and ROI,including the volume related quantitative indicators(the volume of the entire lungs,each pulmonary lobe,double upper lungs and double lower lungs,the total volume of each ROI,the volume and proportion of each ROI at each pulmonary lobe,as well as at the double upper lungs and double lower lungs,and the share of the volume of each ROI in the total volume of the entire lungs,Volume and proportion of pulmonary fibrosis in the upper lungs,lower lungs and the whole lungs,and its volume and proportion in each lung lobe),density related quantitative indicators(the mean density of the entire lungs,of each pulmonary lobes,and of the left and the right lung,the mean density of each ROI,and the mean density of ROI at each pulmonary lobe),etc.SPSS26.0 statistical software was utilized to conduct a contrast analysis of the differences in terms of CT quantitative indicators.The correlation of CT quantitative indicators with pulmonary function parameter and CPI was assessed,and clinical predicting prognosis indicators that are likely to make implications were thereupon picked out.3.Radiomics:OnekeyAI platform was used to perform traditional omics tasks,which include:①Image pre-processing;②Cellular ROI delineation using medical image information system;③image features extracting through OnekeyAI platform software,which mainly include shape features,first-order features and texture features;④Feature dimensional reduction:statistical test(T test or Mann-Whitney U test),feature correlation test(Pearson test),Lasso regression dimensional reduction;⑤Patients were randomly divided at the ratio of 7:3,and the training set and test set were constructed;⑥Model establishing:three algorithms:k-Nearest Neighbor(kNN),Support Vector Machine(SVM)and Light Gradient Boosting Machine(Light GBM)were adopted to establish models in the training set,preserving the best model;⑦Model efficacy test:model efficacy was tested in the test set,including accuracy,sensitivity,specificity,NPV,PPV,F1 score,etc.And the ROC curve was made to calculate AUC.The clinical effectiveness of radiomics models in predicting IPF prognosis.Results:1.Clinical data analysis the research includes a total of 171 IPF cases,among which 134(78.36%)are male and 37(21.64%)female.The mean age of all included patients is 68.95±9.42.101(59.06%)of all cases claimed having smoking history,while 70(40.94%)denied.The mean smoking index is 722.28±565.35.The mean BMI value is 23.32±3.08 kg/m2.73(42.69%)cases admitted having history of alcohol intake,while 98(57.31%)denied.8(4.68%)cases have history of exposure to special essence,while 163(95.32%)denied.As per the dividing standards,the group of favorable prognosis finally included 70 IPF cases,while the one of unfavorable prognosis contained 101 IPF cases.The BMI differences between the two groups have implications statistically(P<0.05).Differences of general information such as sex,age,smoking history,smoking index,history of alcohol intake,and history of exposure to special essence,have no statistical implications(P>0.05).A total of 79 pulmonary function cases were collected,and differences of FVC,FVCpred%,FEV1,FEV1pred%,DLco,DLcopred%,DLco/VA,DLco/VA pred%,TLCpred%,VCpred%,CPI1 and CPI2 in the two groups also have statistical implications(P<0.05).2.CT quantitative analysis ①Quantitative analysis was performed on the volumes of the whole lung,upper lungs,lower lungs,left and right lungs and each lung lobe.There were statistically significant differences in left lower lung volume,right upper lung volume and right lower lung volume(P<0.05).The density-related quantitative indicators of the two groups had significant differences in the mean density of the whole lung,the mean density of the left lung,the mean density of the right lung,the mean density of the right upper lobe,the mean density of the right middle lobe,the mean density of the right lower lobe,the mean density of the left upper lobe,and the mean density of the left lower lobe.Statistical significance(P<0.001);correlation analysis:total lung volume,left lung volume,right lung volume,upper lung volume,lower lung volume,each lung lobe volume and FVC,FVCpred%,TLC,TLCpred%,DLco(except Right upper lung volume quantitative indicators)were positively correlated(P<0.01),and they were negatively correlated with CPI2(/P<0.05),the mean density of the whole lung,the mean density of each lung lobe was negatively correlated with FVC,FVCpred%,TLC,TLCpred%,DLco,DLcopred%(P<0.01);②Quantitative indicators of cellular shadow CT.There were statistically significant differences between the two groups in all cellular volume-related CT quantitative indicators(P<0.05).Among the density-related CT quantitative indicators,the two groups except the mean density of cellular shadow,the mean density in the right middle lung,the mean density in the right lower lung,and the mean density in the left lower lung.There was no statistically significant difference in mean density in the lungs(P>0.05),and other density-related quantitative indicators had statistically significant differences between the two groups(P<0.05);correlation analysis:density-related quantitative indicators and FVC,FVCpred%,TLC,TLCpred%were negatively correlated(P<0.05),volume-related CT quantitative indicators were negatively correlated with DLco,DLcopred%,DLco/VA,DLco/VApred%(P<0.05),and positively correlated with CPI1(P<0.01),other volume-related quantitative indicators except the volume in the right upper lung were positively correlated with CPI2(P<0.05);③Quantitative index of grid shadow CT.There were statistically significant differences between the two groups in all grid shadow density-related CT quantitative indicators(P<0.001);correlation analysis:grid shadow density-related CT quantitative indicators were correlated with FVC,FVCpred%,DLco,DLcopred%,DLco/VA,TLC,and TLCpred%were all negatively correlated(P<0.05),and positively correlated with CPI1 and CPI2(P<0.05),and the proportion of grid shadows in the whole lung was negatively correlated with FVC,FVCpred%,and TLC(P<0.05),positively correlated with CPI1 and CPI2(P<0.05);④CT quantitative index of bronchiectasis shadow.There were statistically significant differences between groups in quantitative indicators such as the total volume of bronchiectasis,the volume in the upper lung,and the proportion of whole lung volume(P<0.05);correlation analysis:the proportion of total lung volume of bronchiectasis and FVC,FVCpred%,TLC,TLCpred%,were negatively correlated(P<0.05),and positively correlated with CPI1,CPI2(P<0.05);⑤Quantitative indicators of ground-glass opacity CT.Between the two groups,there was a statistically significant difference in the CT quantitative indicators of the volume of the middle lung above the ground-glass opacity and the proportion of the mid-lung volume above the ground-glass opacity(P<0.05);correlation analysis:the CT quantitative index of the ground-glass opacity There is no good correlation with lung function parameters(P>0.05).⑥CT quantitative index of pulmonary fibrosis.The differences between the two groups between all volume-related CT quantitative indicators are statistically significant(P<0.05);correlation analysis:volume-related CT quantitative indicators in the upper lung,lower lung,total lung and lobes in the pulmonary fibrosis area are equal to DLco,DLcopred%,DLco/VA,DLco/VApred%The performance is negatively correlated(P<0.001),with a positive correlation with CPI1 and CPI2(P<0.05);the volume proportion of the pulmonary fibrosis area in the whole lung and upper lung CT quantitative indexes are negatively correlated with all lung function indicators(P<0.001),and CPI1 and CPI2.Positive correlation(P<0.001),volume proportion in the lower lung.Except for no correlation with TLC,CT quantitative indexes are negatively correlated with other lung function indicators(P<0.001),positive correlated with CPI1 and CPI2(P<0.001),and the volume proportion of pulmonary fibrosis area in each lung lobes.CT quantitative indicators are negatively correlated with all pulmonary function indicators except TLC(P<0.05)(where the volume proportion of pulmonary fibrosis in the right lower lung CT quantitative index is not correlated with FVC)and positively correlated with CPI1 and CPI2(P<0.001).3.CT radiomics.171 IPF patients were randomly divided at the ratio of 7:3 into the training set(n=119)and test set(n=52).Differences of general information such as sex,age,smoking history,smoking index,history of alcohol intake,and history of exposure to special essence,have no statistical implications(P>0.05).10 radiomics features were filtered out based on cellular shadow ROI.In testing set,the AUC of the model established by means of SVM is 0.897(95%CI:0.810-0.984),the accuracy is 75.000%,the specificity is 85.714%and the sensitivity is 80.645%.The PPV is 71.429%,the NPV is 90.000%,and the F1 score is 0.758.The AUC of the model established by means of kNN is 0.685(95%CI:0.540-0.830),the accuracy is 61.539%,the specificity is 28.571%and the sensitivity is 100%.The PPV is 67.742%,the NPV is 52.381%,and the F1 score is 0.808.The AUC of the model established by means of LightGBM is 0.867(95%CI:0.761-0.973),the accuracy is 78.846%,the specificity is 80.952%and the sensitivity is 90.323%.The PPV is 76.316%,the NPV is 85.714%,and the F1 score is 0.827.Conclusion:1.①Quantitative indicators obtained through CT quantitative analysis:lung volume and average density,honeycomb shadow volume and volume ratio,grid shadow average density,pulmonary fibrosis area volume and volume ratio,etc.There are significant differences between groups,and there is a good correlation with lung function,CPI,and other parameters,which can be used as a new indicator to evaluate the physiological injury or prognosis of IPF patients;②Although there is no significant difference in CT quantitative indicators of mean density of honeycomb shadow between groups,there is a correlation between them and FVC,FVCpred%,TLC,and TLCpred%,which has the reference value to reflect the prognosis;③There is no significant difference in the CT quantitative indicators of the proportion of total lung volume of grid shadow between groups,but there is a correlation with FVC,FVCpred%,CPI1,and CPI2,which has the reference value to reflect the prognosis.④There is a difference in the proportion of total lung volume of bronchodilator shadow between groups in CT quantitative indicators,and there is a certain correlation with FVC,FVCpred%,CPI1,and CPI2,which has the reference value to reflect the prognosis.2.The prediction model established based on CT radiomics has good prediction performance in predicting the prognosis of IPF patients,and has important value in evaluating the prognosis of IPF. |
PDF Full Text Request