Pleural line and B-lines based image analysis for severity evaluation of COVID-19 pneumonia

This paper proposes a quantitative analysis method for lung ultrasound (LUS) images to evaluate the severity of COVID-19 pneumonia. Specifically, biomarkers related to the pleural line, including the thickness of pleural line (TPL) and the roughness of pleural line (RPL), and biomarkers related to the B-lines, including the accumulated width of B-lines (AWBL) and the acoustic coefficient of B-lines (ACBL), are extracted from LUS images to characterize the image patterns associated with the disease severity. 27 patients of COVID-19 pneumonia are enrolled in this study, including 13 moderate cases, 7 severe cases, and 7 critical cases. Patients of moderate cases are regarded as non-severe patients, and patients of severe and critical cases are regarded as non-severe patients. Biomarkers among different cases are compared, and the performances in the binary diagnosis of severe and non-severe patients are assessed using a support vector machine (SVM) classifier with all the biomarkers as the input. The classification performance is optimal using the SVM classifier (area under the receiver operating characteristics curve = 0.93, sensitivity = 0.93, specificity = 0.85). The proposed method may be a promising tool for the automatic grading and follow-up of patients with COVID-19 pneumonia.

Automated summary

This paper introduces a quantitative analysis method for evaluating the severity of COVID-19 pneumonia using lung ultrasound (LUS) images. By extracting biomarkers related to pleural line and B-lines from LUS images, patients were classified into different severity levels, and a SVM classifier was used to assess the binary diagnosis performance. The optimal classification performance indicates that the proposed method may be a promising tool for automatic grading and follow-up of patients with COVID-19 pneumonia.