Study On The Construction Of Predictive Model Of Refractory Mycoplasma Pneumoniae Pneumonia In Children By AI Quantitative Analysis Based On CT Images

Objective:In view of the lack of effective methods for early diagnosis of refractory mycoplasma pneumoniae pneumonia(RMPP)in children,this study makes a quantitative analysis of children’s refractory mycoplasma pneumoniae pneumonia based on AI of CT images,looked for sensitive indexes of imaging quantitative analysis,studied the diagnostic value of imaging quantitative indexes for RMPP,and discussed the related predictive factors of predicting RMPP in combination with clinical practice,and established a Nomogram risk prediction model to provided basis for early clinical diagnosis and treatment.This study is divided into three parts:Part one:To verify the accuracy of AI in the detection of pulmonary lesions in children with mycoplasma pneumonia,and to explore the correlation between CT quantitative index and clinical laboratory examination index.Part two:To analyze the characteristics of quantitative CT index in children with RMPP based on AI,and to explore the diagnostic efficacy of quantitative CT index in RMPP.Part three:Study on the construction of predictive model of refractory pneumonia in children based on AI,combined with clinical data to find predictive indicators for early identification of children with RMPP,and establish an early predictive model for children with RMPP.Part one1.Materials and methods51 children with mycoplasma pneumonia were included in the affiliated Hospital of Yan’an University.The original data of CT scan were imported into AI analysis software in"Dicom"format for segmentation processing,and the lesion parameters included Lesion volume,Lesion volume percentage,Mean lesion density,Ground glass opacity percentage and Lesion weight.Pearson or Spearman correlation analysis was used to analyze the correlation between the lesion range score evaluated by physicians and the Lesion volume percentage obtained by AI,the lesion density score evaluated by physician and the Mean lesion density obtained by AI,the correlation between the total lesion score evaluated by physician and the Lesion weight obtained by AI,and the correlation between Lesion volume percentage,Mean lesion density and Lesion weight obtained by AI and laboratory examination.2.ResultsThe score of lesion range evaluated by physicians was highly correlated with the Lesion volume percentage(r=0.855 P<0.01),the score of lesion density was highly correlated with the Mean lesion density(r=0.743 P<0.01),and the score of lesion severity was highly correlated with the Lesion weight(r=0.903 P<0.01).The Lesion volume percentage was slightly positively correlated with the percentage of neutrophils and C-reactive protein,and slightly negatively correlated with the percentage of lymphocytes and prealbumin,and the Mean lesion density was slightly positively correlated with C-reactive protein and slightly negatively correlated with the percentage of lymphocytes.The Lesion weight was mildly to moderately positively correlated with neutrophil percentage and C-reactive protein(P<0.05),and negatively correlated with lymphocyte percentage(P<0.05).3.Brief summaryIn this study,the results of AI quantitative analysis were compared with the results of manual evaluation by radiologists,and it was found that AI quantitative analysis had a high correlation with the scope and severity of lesions compared with physicians,and mild to moderate correlation with laboratory examination.It is an objective,accurate and feasible automatic evaluation method.Part two1.Materials and methods197 children with MPP were enrolled in the affiliated Hospital of Yan’an University,the second Hospital of Yulin City and Xingyuan Hospital of Yulin,including 137 children with general mycoplasma pneumonia(GMPP)and 60 children with RMPP.The diagnostic criteria of RMPP are based on persistent fever,clinical manifestations and imaging deterioration after treatment with macrolide antibiotics for 7 days or more.The measurement data conforms to the normal distribution by X±s,the independent sample T test is used for the comparison between the two groups,the M(IQR)is used for the non-normal distribution,the Mann-Whitney U test is used for the comparison between the two groups,the counting data is expressed by n(%),and the chi-square test is used for the comparison between the two groups.The area under the curve(AUC)of receiver operating characteristics curve(ROC)was used to analyze the diagnostic value of quantitative CT index in RMPP.The difference was statistically significant(P<0.05).2.ResultsThe age of children with mycoplasma pneumonia in RMPP group was older than that in GMPP group(Z=-2.373,P<0.05),and the average hospital stay in GMPP group was longer than that in GMPP group(Z=-8.058,P<0.05).The fever rate in the RMPP group was higher than that in the GMPP group(X~2=12.631,P<0.05),but there was no significant difference in the incidence of cough between the two groups(X~2=0.896,P>0.05).In laboratory examination,there was no significant difference in white blood cells between the two groups(Z=0.524 P>0.05).Neutrophil counts and percentage(Z=-3.372,t=-5.652),,C-reactive protein(Z=-5.509)and lactate dehydrogenase(t=-3.939)in RMPP group were significantly higher than those in the GMPP group(all P<0.05);The lymphocyte count and percentage(Z=-3.748,t=5.601),monocyte counts and percentage(Z=--1.019,t=-2.270)and the level of prealbumin(Z=-2.332)in the RMPP group were lower than those in the GMPP group(all P<0.05);There was no significant difference in gender(X~2=0.139),respiratory virus mixed infection rate(X~2=0.090),platelet count(Z=-0.522)and immunoglobulin(statistical value=-1.472～1.248)between the two groups Significance(all P<0.05).There were significant differences in quantitative CT between RMPP group and GMPP group(Z=-8.629～-5.761,P<0.05).Compared with children with GMPP,Le V,Le V%,MLe D and LM were higher in children with RMPP,while GGO%was lower in children with RMPP.Le V,Le V%,MLe D,GGO%and LM can distinguish children with GMPP from children with RMPP.When Le V>22.9ml,Le V%>5.01%,MLe D>-296.36HU,GGO%<0.47,LM>20.8,the areas under the curve classified as RMPP were 0.861,0.868,0.679,0.758 and 0.887,respectively.3.Brief summaryIn this study,CT quantitative analysis technique was used to objectively and quantitatively determine the lesion volume,lesion proportion,mean lesion density,ground glass opacity percentage of the lesion and lesion quality in the lungs of children with RMPP.The results showed that the lesion range was wide,the lesion density was high,the lesion quality was high,and the proportion of ground glass density lesions was low.When the volume of the whole lung lesion reaches 22.9ml,the proportion of the lesion reaches 5.01%,the density of the lesion reaches-296.36HU,the proportion of ground glass density reaches0.47%,and the mass of the lesion reaches 20.8 g,RMPP children can be better identified.Part three1.Materials and methodsThe included object is the same as the second part.Among them,the children were collected from the affiliated Hospital of Yan’an University as the training set,and the second Hospital of Yulin City and Xingyuan Hospital of Yulin as the verification set.The measurement data conforms to the normal distribution by X±s,the independent sample T test is used for the comparison between the two groups,the M(IQR)is used for the non-normal distribution,the Mann-Whitney U test is used for the comparison between the two groups,the counting data is expressed by n(%),and the chi-square test is used for the comparison between the two groups.Multivariate stepwise logistic regression was used to screen predictive factors,and the selected factors were used to establish multivariate logistic regression models.In this study,three predictive models were established,namely,a clinical model using only clinical features,an imaging model using only quantitative CT indicators,and a clinical-imaging model using both clinical and quantitative CT.ROC’s AUC was used to evaluate the discrimination ability of the model in the training set and verification set.Finally,the optimal model is represented by nomogram,and the accuracy of nomogram is evaluated by the discriminant ability and calibration map of training set and verification set.Through the Hosmer-Lemesshow test,the calibration chart is established to verify the model.The clinical validity of nomogram was evaluated by decision curve analysis(DCA).The difference was statistically significant(P<0.05).2.ResultsA total of 197 children with mycoplasma pneumonia were included in this study,including 33 cases of RMPP and 83 cases of GMPP,and 81 cases of RMPP and 54 cases of GMPP.There was no significant difference between clinical data and CT quantitative indexes except lymphocyte percentage(Z=-2.080)and platelet count(Z=-2.266)in training set and verification set.There were significant differences in hospitalization time,fever rate,neutrophil count and percentage,monocyte count and percentage,CRP,LDH,lymphocyte count and percentage and prealbumin between RMPP group and GMPP group.Multivariate stepwise logistic regression analysis showed that CRP,LDH,Le V%and GGO%were independent predictors of RMPP.Based on the above four predictive factors,three kinds of regression models for predicting RMPP were established.Through the analysis of ROC curve,it was concluded that the clinical-imaging quantitative model had the highest AUC,which was 0.933 and 0.922 in training set and verification set respectively,which was the best model and represented by nomogram diagram.The calibration curve shows that the predicted values of the training set and the verification set are basically consistent with the measured values.Hosmer-Lemesshow test shows that there is no significant difference between the model and the ideal model in the training set and verification set,indicating that the model has good accuracy in predicting RMPP.DCA curve shows that the model has good clinical application value.3.Brief summaryIn this study,a nomogram risk prediction model for children with RMPP was developed and verified by introducing imaging quantitative indexes,clinical features and laboratory examination indexes for the first time.C-reactive protein,lactate dehydrogenase,lesion proportion and ground glass density were included in the model.Nomogram risk prediction model shows good accuracy and differentiation,suggesting that it can help clinicians to identify refractory mycoplasma pneumonia in children earlier,so as to intervene and treat as soon as possible.ConclusionsBy comparing the results of quantitative CT with the results of manual evaluation by radiologists,it is found that quantitative CT has a high consistency in the scope and severity of lesions compared with physicians,and has a certain correlation with laboratory examination,so it is an objective,accurate and feasible automatic evaluation method.Quantitative CT technique was used to quantify the volume,proportion,density,proportion of ground glass density and mass of pulmonary lesions.The results showed that when the volume of whole lung lesions reached 22.9ml,the proportion of lesions reached5.01,the density of lesions reached-296.36HU,the proportion of ground glass density was less than 0.47,and the mass of lesions reached 20.8g,RMPP could be well identified.A Nomogram risk prediction model was developed and verified by imaging quantitative indexes combined with clinical features and laboratory examination indexes.C-reactive protein,lactate dehydrogenase,lesion proportion and ground glass density ratio were included in the model,which showed good accuracy and differentiation.It is helpful for early clinical identification of refractory mycoplasma pneumonia in children,so as to intervene and treat as soon as possible.