Background
Some key measures to control COVID-19 and to provide adequate patient care are fast and reliable diagnosis and severity assessment. Of the initially moderate to severe cases, some recover completely while others deteriorate and need invasive ventilation support. The long-term effects on the lung in severe cases are largely unknown. The predictive role of radiology and laboratory and clinical parameters for the acute and chronic state is not yet well understood.
Research aims
Together with an international team from Inselspital Bern, the University of Parma, Italy, and Yale University, USA, we wish to develop an artificial intelligence (AI)-based system that combines information from thoracic computer tomography, laboratory parameters and clinical data to predict the seven-day progression based on the current condition of a patient in the acute phase as well as the chronicity of the disease. This will allow better assessment of the current condition of the patient, identification of patients at risk and could ultimately provide important information for planning hospital resources under intense pressure.
Expected results and envisaged products
Due to the large database and data depth as well as the integrative design, we expect our multi-omics AI-based approach to achieve higher accuracy than an AI-based model based solely on imaging or non-imaging information. This model is also intended to contribute to a better understanding of the various factors relevant for assessing and predicting the severity of COVID-19. As soon as the AI model is established, we plan to develop a COVID-19 app for risk stratification with high diagnostic accuracy.
Specific contribution to tackle the current pandemic
Determining faster and more accurate diagnoses as well as assessment of expected disease outcome in COVID-19 patients based on the proposed AI system will contribute to better risk assessment of the disease and improve physician efficiency. In particular, we expect radiologists to be able to make faster and more accurate diagnoses. Rapid implementation of our system will facilitate patient care and improve our understanding of which parameters or which combination of parameters is relevant to deterioration of the patient’s condition.
COVID-19 is a pandemic disease with tremendous consequences spreading at fast pace. Key measures to control and manage this currently untreatable disease, and to provide adequate patient care are rapid and reliable diagnosis as well as severity assessment. While most patients develop only mild symptoms or are asymptomatic altogether, others exhibit severe courses that are associated with a high mortality rate. Of the initially moderate to severe cases, some recover completely while others deteriorate with the need of being put on invasive ventilation support or even extracorporeal membrane oxygenation. The long term effects on the lung in severe cases are largely unknown to date. First publications suggest that pulmonary fibrosis could develop. In addition, over the last weeks we have learned that the prognosis depends not only on pathological changes in lung tissue but also on changes in a variety of other pathological sites that the virus attacks such as the vessel wall. The findings on chest imaging for COVID-19 are often typical, but ultimately not specific and overlap with other infections, including influenza, H1N1. Despite early positive reports, the role of radiology in the management of COVID-19 remains to be defined. In this project we aim to develop and test an AI-based multi-omics system that combines and uses information from chest CT, laboratory parameters, and clinical data to a) assess the current state of a patient in the acute phase and to forecast seven-day progression and b) to predict chronicity (chronic lung damage). More specifically, our research is aimed at investigating whether an ensemble of multi-omic, patient-specific information can be used to predict patient outcome and ultimately optimize patient care. In addition, we intend to interpret and understand the importance of each item in the ensemble in terms of learning and disease outcome prediction. Special emphasis will be placed on the vascular situation and certain lung changes (bronchiectasis). Furthermore, through the predictive capabilities of the proposed AI-based multi-omics approach, we aim to better understand how COVID-19 progresses over time, since it is largely unknown why patients differ in disease progression. Finally, once the AI system is established, we intend to monitor treatment response in patients receiving different treatments (e.g. Remdesivir, Actrema). To date, we have access to more than 2’000 chest CTs of patients with laboratory-proven COVID-19 infection as well as their lab parameters, age, gender, and patient history. The majority of COVID-19 positive cases will be provided from centers in Northern Italy. As controls, we will include 1’200 cases with similar symptoms who have had pathological CT findings (pneumonias of various causes) before 12/2019 to rule out that these symptoms could have been caused by COVID-19 and 1’000 negative controls with normal chest CT. We will measure the performance of the AI-based computational engine, as well as the contribution of each individual variable to the classification and prediction performance of the proposed system. Quantitative metrics used to analyze the results include sensitivity, specificity, accuracy, positive predictive value and area under the curve (AUC) of the Receiver Operating Characteristic (ROC).Determining faster and more accurate prognostic stratification for COVID-19 patients based on the proposed AI-system is expected to contribute to better and more appropriate patient care and improve the efficiency of physicians. Better understanding of the different parameters and their interaction for the outcome of COVID-19 patients can shed new insights regarding the disease progression patterns of COVID-19, which could be of crucial importance for the treatment of critical patients.
Last updated:18.07.2023
