| Part Ⅰ Artificial intelligence-assisted CT quantitative study of pulmonary lesions in COVID-19 patientsObjective: The purpose of this study was to explore the dynamic change of COVID-19 lesions during the acute phase and convalescent phase by quantitative indexes obtained with the assistance of artificial intelligence(AI)from sequential chest CT scans.Furthermore,an interpretable machine learning model was used to predict the presence of residual pneumonia lesion in the convalescent period.Methods: Moderate and severe COVID-19 patients who were hospitalized in our hospital from January 10,2020 to April 3,2020 were enrolled.The u AI Intelligent Assistant Analysis System(United Imaging Medical Technology Company Limited)was used to obtain the quantitative CT indexes of the total lung and five lobes in acute COVID-19phase(at admission and discharge)and convalescent phase(at 3-month and 6-month after discharge).The well-aerated lung(WAL)was considered as lung tissue with CT values between-950 HU and-750 HU.Different components of the lesions were distinguished based on the CT values as follows,ground-glass opacity(GGO,-750 HU ~-300HU);solid components(SC,-300 HU ~ 50HU).A total of 14 CT quantitative indexes reflecting lesion volume,lesion composition ratio and total lung or lobar volume were recorded respectively.These indexes were used to describe the dynamic change of pneumonia lesions in moderate and severe COVID-19 patients,and the Mann-Whitney U test was used to compare the differences in the lesion composition at each time point between the two groups.Then set the endpoint as whether the patient with residual lung lesions at3-months after discharge.The e Xtreme Gradient Boosting(XGBoost)machine learning method was employed to construct the prediction models respectively and compared:admission model,discharge model.Interpretation of the prediction models were performed using SHapley Additive ex Planations(SHAP),which can effectively and accurately calculate the contribution of features to the model.Results: A total of 136 COVID-19 patients were enrolled,77(56.6%)in the moderate group,and 59(43.4%)in the severe group.Regardless of clinical type,there were no differences in total lung lesion volume(Le V)at discharge compared with at admission(P= 0.973 and 0.421,respectively),but the actual lesion volume(ALe V,which represents the Le V excluding WAL within the infected region)was significantly reduced in the severe group(at admission vs at discharge: 388.9 [136.8,647.3] cm3 vs 162.8 [58.7,431.5] cm3,P < 0.001).At 3-month after discharge,68(50%)patients had residual lesion,and Le V was significantly reduced in both groups when compared with acute phase(moderate: 0 [0,11.95] cm3;severe: 20.1 [0,96.5] cm3).Severe patients were more likely to have residual lesion,most of which was located in the right lower lobe(RLL).At 6-month after discharge,the median of each lesion index decreased to 0 cm3.From the analysis of the composition of lesions,GGO was the main component in the acute phase,and WAL in the convalescent phase.The proportion of SC in Le V in the severe group(SCVto Le%)at admission was higher than that in the moderate group(severe vs moderate: 17.98 [9.02,28.27] % vs 7.43 [0.71,15.35] %,P<0.001).However,there were no differences in the composition of residual lesions after discharge between two group(all P>0.05).Comparing the XGBoost models constructed from the total lung CT quantitative indexes at admission and discharge,it was showed that discharge CT model had better predictive performance than admission CT model(AUC of discharge vs admission model: 0.9123 vs0.8099).Furthermore,the model based on the lobar quantitative indexes at discharge performed better than that based on the total lung indexes(AUC of lobar vs of total lung model: 0.9439 vs 0.9123).The SHAP analysis showed CT quantitative indexes of the left upper lobe(LUL)and the RLL had high influence on the prediction performance and the most important feature was ALe VLUL.Conclusions: AI-assisted CT quantitative analysis at the total lung and lobar level could provide objective information on COVID-19.The XGBoost machine learning model suggested a greater predictive value of CT indexes at discharge for residual lesions at3-month after discharge,which is helpful for clinical formulating targeted diagnosis and treatment and follow-up strategies for patients.Part Ⅱ Quantitative study of pulmonary function in COVID-19 patients based on paired inspiratory–expiratory chest CT scanObjective: The purpose of this study was to prove the similarity and difference among normal pulmonary lobes in the respiratory process through the quantitative CT indexes,and then to explore the application value of paired inspiratory–expiratory chest CT scan in pulmonary function of convalescent COVID-19 patients.Methods: COVID-19 patients admitted to our hospital from January 10,2020 to April 3,2020 were enrolled and followed up with clinical pulmonary function test(PFT)and paired inspiratory–expiratory chest CT scan at 3-month after discharge.Age-and sex-matched normal subjects who underwent the above-mentioned two tests were also included as the control group.The COPD Analysis post-processing software(Philips Intelli Space Portal,version 12.0)was used to obtain CT quantitative indexes of the total lung and each lobe.The Friedman test was used to compare the CT quantitative indexes in the inspiratory and expiratory phases of the five normal lobes(left upper lobe [LUL],left lower lobe [LLL],right upper lobe [RUL],right middle lobe [RML] and right lower lobe[RLL]).Spearman correlation analysis and multiple linear regression were used to analyze the relationship between CT quantitative indexes(LVin-TL,WALTL,LVex-TL,?LVTL)and PFT indexes(TLC,RV,FVC)of normal subjects who completed these two examinations within 3 days.According to PFT results in convalescence of COVID-19 patients in convalescence,they were divided into normal ventilatory function group and ventilatory dysfunction group;normal diffusion function group,diffusion dysfunction group.Mann-whitney U test was used to compare the CT quantitative indexes of the total lung and 5 lobes of patients with the control group.Results: A total of 104 COVID-19 patients were enrolled,the PFT results showed 48(46.2%)patients with ventilatory dysfunction(obstructive,restrictive and mixed ventilatory dysfunction were 15 [14.4%],17 [16.3%] and 16 [15.4%],respectively).There were 38(36.5%)patients with diffusion dysfunction.A total of 70 normal subjects with PFT and paired inspiratory–expiratory CT scan were collected,65 of whom completed both tests within 3 days.The lobar volume and mean lung density of the five lobes were not demonstrated the same in both inspiratory and expiratory phases(all P<0.001).The bilateral lower lobes presented larger volume change(?LVLLL was 604.35 [503.8,727.64]cm3,?LVRLL was 668.99 [548.22,773.07] cm3),and the ratio of lobar volume to the total lung volume(λin and λex)indicated that the bilateral lower lobes were responsible for more ventilation than it should proportionately take(the λin-LLL and λin-RLL were higher thanλex-LLL and λex-RLL,respectively,both P<0.001).There presented good correlation between the CT quantitative indexes and PFT indexes: LVin-TL and WALTL were positively correlated with TLC(r=0.890 and 0.879,respectively).LVex-TL was positively correlated with RV(r=0.811).WALTL showed a better correlation than ?LVTL with FVC(r=0.817 and 0.719,respectively),and these two regression equation established by lobar indexes showed better fit(WAL equation: R2=0.677;?LV equation: R2=0.576).Analysis of ventilatory function in COVID-19 patients in convalescence showed that ventilatory status was not associated with clinical type of COVID-19(moderate or severe),P=0.362.But the proportion of residual lesion in ventilatory dysfunction group was higher than that in normal ventilatory function group(62.5% vs 41.1%,P=0.029).The ΔLVTL and WALTL%were lower than controls regardless of normal ventilation function(all P<0.05).For the normal ventilatory function group,the abnormal CT indexes were mainly located in the bilateral lower lobes;while for the ventilatory dysfunction group,the abnormal CT indexes were mainly located in the RLL.Analysis of diffusion function of COVID-19 patients in convalescence showed that the status of diffusion function was not related to residual lesions(P=0.139),but the proportion of severe patients in the diffusion dysfunction group was higher than that in the normal diffusion function group(68.4% vs47.0%,P=0.034).In normal diffusion function group,only the ΔLVTL lower than that in the control group(2272.35 [1758.9,2804.35] cm3 vs 2553.05 [2183.15,3102.71] cm3,P=0.034),and the abnormal CT values were mainly located in RLL.In the diffusion dysfunction group,only LVex-TL showed no difference with the control group(P=0.363), the other four indexes were all lower than those in the control group(LVin-TL,?LVTL,WALTL,WALTL%,all P<0.05),abnormal CT values were mainly from LUL,LLL and RLL.In addition,there were no differences between the CT quantitative indexes of RML and the control group,regardless of the pulmonary function status in convalescence(all P > 0.05).Conclusions: The 5 normal lobes were non-synchronous during respiration and contributed differently to ventilation.The bilateral lower lobes showed similarities and had a high-ventilation function.There is a good correlation between the quantitative indexes of paired inspiratory–expiratory chest CT and PFT.Lobar-based CT quantitative analysis enables greater detail of COVID-19 patients under different pulmonary function states in the convalescence,which demonstrated the feasibility to further refine the evaluate clinical pulmonary function. |
PDF Full Text Request