Research On Early Identification Of Pulmonary Necrosis And Prediction Of Bronchiectasis In Children Based On Radiomics And Machine Learnin

Objective:Necrotizing pneumonia(NP)is an infrequent but severe complication of pneumonia in children.In the early stages of NP,CT imaging shows lung consolidation,which cannot be detected in time.Necrotizing pneumonia can be diagnosed with enhanced CT or follow-up,but with increased doses of ionizing radiation and iodine contrast agent intake.This study aimed to explore the ability of non-contrast-enhanced CT radiomics features torecognize necrotizing pneumonia in early stage.Methods: This was a retrospective study,and 250 patients who presented with lung consolidation on initial CT images were included in this study.After a follow-up period of 1–3 weeks,116 patients developed NP,whose CT or X-ray shows cavitation or liquefied necrosis.Manual segmentation of lesion sites in the initial non-contrast-enhanced CT scans was performed with Rad Cloud(Huiying Medical Technology Co.,Ltd.China),and 1409 radiomics features were extracted.We used Variance threshold(0.8),Select KBest,and the least absolute shrinkage and selection operator(LASSO)methods for feature dimension reduction.Three machine learning algorithms,k-nearest neighbor(KNN),support vector machine(SVM),and logistic regression(LR)models,were established to recongnize NP early.To assess the recongnition performance,the area under the receiver operating characteristic curve(AUC),sensitivity,specificity,and other indicators were used in the validation cohort.Results: 1.The regions of interest(ROIs)of lesion were manually segmented on CT images,and then 1409 radiomics features were extracted using the Rad Cloud platform.2.The variance threshold(0.8)method was used to preliminarily select 531 features from 1409 features,and then 297 features were identified by univariate selection method.Finally,Lasso algorithm was used for dimensionality reduction,and 15 optimal radiomics features were finally determined.3.Radiomics features helped to recongnize NP in early stage in both the training and validation cohorts.The AUC(sensitivity,specificity)for the training and validation cohorts were 0.81(0.73,0.68)and 0.71(0.61,0.65)for KNN,respectively;0.81(0.72,0.70)and 0.77(0.66,0.65)for SVM,respectively;and0.82(0.73,0.73)and 0.76(0.63,0.70)for LR,respectively.Recall and F1-scores determined that LR performed better at diagnosing early NP,with the values of the above two indexes being 0.70 and 0.67,respectively.Conclusion: Radiomics features extracted from the region of interest(ROI)of lung consolidation are correlated with the occurrence of children’s necrotizing pneumonia,and the establishment of a child’s machine learning model based on radiomics features is helpful for early recognition of NP.Objective: Pulmonary infection is the most common cause of bronchiectasis in children.The purpose of this study was to investigate the ability of radiomics features with machine learning based on non-contrast enhanced CT images to predict bronchiectasis in children after pulmonary infection.Materials and methods: In this study,57 patients with pulmonary consolidation who developed bronchiectasis in a period of time were included according to the inclusion and exclusion criterion.According to the presence or absence of bronchiectasis,the cohort was divided into two groups: bronchiectasis group(n=19)and non-bronchiectasis group(n=38).CT radiographic and clinical data of enrolled patients were collected.The ROIs of lesion sites were manually segmented on Rad Cloud platform,and then radiomics features were extracted automatically.The variance threshold method,select K best method and minimum absolute contraction and selection operator method were used to features dimensionality reduction.Two machine learning algorithms,K-nearest neighbor(KNN)and Logistic regression(LR),were used to build predictive models.Area under receiver operating characteristic curve(AUC),sensitivity,specificity and other indicators were used to evaluate the predictive performance in the validation cohort.Results: The ROIs of lesions in the CT images were manually segmented,and then 1409 radiomics features were automatically extracted using the Rad Cloud platform.The variance threshold(0.8)method was used to select 451 features from 1409 features,and 35 features were identified by select K best method.Finally,9 radiomics features that were strongly correlated with bronchiectasis were determined by LASSO method.Two machine learning algorithms,KNN and LR,were used to establish the predictive models.The AUC and F1-scores of KNN in the training set were 0.93 and 0.83,respectively,while those of KNN in the validation set were 0.84 and 0.53,respectively.The AUC and F1-scores of LR in the training set were 0.94 and 0.79,respectively,while those of LR in the test set were 0.64 and 0.63,respectively.Conclusion: Radiomics features extracted from pulmonary consolidation are helpful to predict the occurrence of bronchiectasis after pulmonary infection,and machine learning model established by KNN performs better in predicting bronchiectasis.