A model based on CT radiomic features for predicting RT-PCR becoming negative in coronavirus disease 2019 (COVID-19) patients

BMC Medical Imaging

Quan Cai¹  Gang Hou²  Pei-Ling Li³  Shu Li³  Xin Wang³  Zheng Zhang³  Si Gao³  Si-Yao Du³  Peng Lv³  Guo-Liang Huang³  Li-Na Zhang³ 

[1] Department of Emergency Medicine, The First Affiliated Hospital of China Medical University, Nanjing North Street 155, 110001, Shenyang, Liaoning Province, China;Department of Pulmonary and Critical Care Medicine, The First Affiliated Hospital of China Medical University, 110001, Shenyang, China;Department of Radiology, The First Affiliated Hospital of China Medical University, 110001, Shenyang, China;

关键词: COVID-19;    Computed tomography;    Radiomics;    Quantitative;    RT-PCR;   

DOI  :  10.1186/s12880-020-00521-z

来源: Springer

【 摘 要 】

BackgroundCoronavirus disease 2019 (COVID-19) has emerged as a global pandemic. According to the diagnosis and treatment guidelines of China, negative reverse transcription-polymerase chain reaction (RT-PCR) is the key criterion for discharging COVID-19 patients. However, repeated RT-PCR tests lead to medical waste and prolonged hospital stays for COVID-19 patients during the recovery period. Our purpose is to assess a model based on chest computed tomography (CT) radiomic features and clinical characteristics to predict RT-PCR negativity during clinical treatment.MethodsFrom February 10 to March 10, 2020, 203 mild COVID-19 patients in Fangcang Shelter Hospital were retrospectively included (training: n = 141; testing: n = 62), and clinical characteristics were collected. Lung abnormalities on chest CT images were segmented with a deep learning algorithm. CT quantitative features and radiomic features were automatically extracted. Clinical characteristics and CT quantitative features were compared between RT-PCR-negative and RT-PCR-positive groups. Univariate logistic regression and Spearman correlation analyses identified the strongest features associated with RT-PCR negativity, and a multivariate logistic regression model was established. The diagnostic performance was evaluated for both cohorts.ResultsThe RT-PCR-negative group had a longer time interval from symptom onset to CT exams than the RT-PCR-positive group (median 23 vs. 16 days, p < 0.001). There was no significant difference in the other clinical characteristics or CT quantitative features. In addition to the time interval from symptom onset to CT exams, nine CT radiomic features were selected for the model. ROC curve analysis revealed AUCs of 0.811 and 0.812 for differentiating the RT-PCR-negative group, with sensitivity/specificity of 0.765/0.625 and 0.784/0.600 in the training and testing datasets, respectively.ConclusionThe model combining CT radiomic features and clinical data helped predict RT-PCR negativity during clinical treatment, indicating the proper time for RT-PCR retesting.

【 授权许可】

CC BY