Automated Severity Assessment of COVID-19 based on Clinical and Imaging Data

Coronavirus disease 2019 (COVID-19) has overwhelmed health systems worldwide. It is important to identify severe cases as early as possible, so that resources can be mobilized and treatment can be escalated. The goals of this project are: (1) to develop a machine learning approach for automated severity assessment of COVID-19 patients based on clinical and imaging data; (2) to identify the clinical and imaging features that have high predictive power of patients’ severity.

We instigated collaborations with two universities and two hospitals in China in response to the call for COVID-19 research. Using data from our international partners, we have developed an AI system for automated lung lesion detection, severity assessment and progression prediction. We compared the predictive power of clinical and imaging data using the SHapley Additive exPlanations (SHAP) framework, and examined the data of 346 patients with COVID-19 by testing multiple machine learning models. The results of this project indicate that clinical and imaging features can be used for automated severity assessment of COVID-19 patients. While imaging features had the strongest impact on the severity assessment performance, inclusion of clinical features yielded the better performance. Our proposed method may have the potential to assist with triaging COVID-19 patients and prioritizing care for patients at higher risk of severe cases.