High-Dimensional Multinomial Multiclass Severity Scoring of COVID-19 Pneumonia Using CT Radiomics Features and Machine Learning Algorithms

We aimed to construct a prediction model based on computed tomography (CT) radiomics features to classify COVID-19 patients into severe-, moderate-, mild-, and non-pneumonic. A total of 1110 patients were studied from a publicly available dataset with 4-class severity scoring performed by a radiologist (based on CT images and clinical features). CT scans were preprocessed with bin discretization and resized, followed by segmentation of the entire lung and extraction of radiomics features. We utilized two feature selection algorithms, namely Bagging Random Forest (BRF) and Multivariate Adaptive Regression Splines (MARS), each coupled to a classifier, namely multinomial logistic regression (MLR), to construct multiclass classification models. Subsequently, 10-fold cross-validation with bootstrapping (n=1000) was performed to validate the classification results. The performance of multi-class models was assessed using precision, recall, F1-score, and accuracy based on the 4×4 confusion matrices. In addition, the areas under the receiver operating characteristic (ROC) curve (AUCs) for multi-class classifications were calculated and compared for both models using “multiROC” and “pROC” R packages. Using BRF, 19 radiomics features were selected, 9 from first-order, 6 from GLCM, 1 from GLDM, 1 from shape, 1 from NGTDM, and 1 from GLSZM radiomics features. Ten features were selected using the MARS algorithm, namely 2 from first-order, 1 from GLDM, 2 from GLRLM, 2 from GLSZM, and 3 from GLCM features. The Mean Absolute Deviation and Median from first-order, Small Area Emphasis from GLSZM, and Correlation from GLCM features were selected by both BRF and MARS algorithms. Except for the Inverse Variance feature from GLCM, all selected features by BRF or MARS were significantly associated with four-class outcomes as assessed within MLR (All p-values<0.05). BRF+MLR and MARS+MLR resulted in pseudo-R² prediction performances of 0.295 and 0.256, respectively. Meanwhile, there were no significant differences between the feature selection models when using a likelihood ratio test (p-value =0.319). Based on confusion matrices for BRF+MLR and MARS+MLR algorithms, the precision was 0.861 and 0.825, the recall was 0.844 and 0.793, whereas the accuracy was 0.933 and 0.922, respectively. AUCs (95% CI)) for multi-class classification were 0.823 (0.795-0.852) and 0.816 (0.788-0.844) for BRF+MLR and MARS+MLR algorithms, respectively. Our models based on the utilization of radiomics features, coupled with machine learning, were able to accurately classify patients according to the severity of pneumonia, thus highlighting the potential of this emerging paradigm in the prognostication and management of COVID-19 patients.