Construction and comparison of multiple classification models based on chest CT images and clinical cha-racteristics for predicting myocardial injury in hospitalized patients with COVID-19
Construction and comparison of multiple classification models based on chest CT images and clinical cha-racteristics for predicting myocardial injury in hospitalized patients with COVID-19
Objection:To construct and validate classification models based on chest CT images and clinical characteristics for timely predicting myocardial injury in patients with COVID-19.Methods:We retrospectively collected 382 laboratory-confirmed hospitalized patients with COVID-19 from November 2022 to February 2023 who underwent chest CT.Myocardial injury was defined as high-sensitivity cardiac troponin Ⅰ(Hs-cTnⅠ)levels above 17.5μg/mL.There were 143 patients with myocardial injury and 239 patients without.Clinical date,laboratory results and four quantitative pa-rameters on chest CT images(including ascending aortic diameter,main pulmonary artery diameter,pulmonary inflammation index,and pulmonary infection proportion)were analyzed between the two groups.All patients were randomly assigned to the training set(306 cases)and the validation set(76 cases)according to a ratio of 8∶2.Logistic regression algorithm in machine learning was employed to construct CT model,clinical model and CT-clinical combined model respectively based on quantitative CT parameters,clinical data and their combination for predicting myocardial injury.Additionally,a deep learning(DL)model based on CT images was developed and validated,with heat maps visuali-zing the regions of interest identified by the DL model to assess the model's interpretability and ratio-nality.The predictive efficacy of the three models was compared based on the area under the receiver operation characteristic curve(AUC).Results:The DL model showed good discrimination performance in the validation cohort(AUC=0.970),which was higher than the CT model(AUC=0.735),clinical model(AUC=0.907)and CT-clinical model(AUC=0.920).The heat map showed a good subjective consistency between the image areas considered relevant by the DL model for myocardial injury classi-fication and those areas of interest in clinical diagnosis.Conclusion:The deep learning model based on chest CT images provides a timely,non-invasive,and relatively accurate means for predicting COVID-19 associated myocardial injury,offering additional information to aid clinicians in patient stratification and decision-making.
关键词
新型冠状病毒肺炎/心肌损伤/深度学习/体层摄影术,X线计算机/影像组学/预测模型
Key words
Coronavirus Disease 2019/Myocardial injury/Deep learning/Tomography,X-ray computed/Radiomics/Prediction model
维普
万方数据