Machine Learning based COVID-19 Diagnosis from Blood Tests with Robustness to Domain Shifts

Abstract

We investigate machine learning models that identify COVID-19 positive patients and estimate the mortality risk based on routinely acquired blood tests in a hospital setting. However, during pandemics or new outbreaks, disease and testing characteristics change, thus we face domain shifts. Domain shifts can be caused, e.g., by changes in the disease prevalence (spreading or tested population), by refined RT-PCR testing procedures (taking samples, laboratory), or by virus mutations. Therefore, machine learning models for diagnosing COVID-19 or other diseases may not be reliable and degrade in performance over time. To countermand this effect, we propose methods that first identify domain shifts and then reverse their negative effects on the model performance. Frequent re-training and re-assessment, as well as stronger weighting of more recent samples, keeps model performance and credibility at a high level over time. Our diagnosis models are constructed and tested on large-scale data sets, steadily adapt to observed domain shifts, and maintain high ROC AUC values along pandemics.

Applications of Machine Learning and High-Performance Computing in the Era of COVID-19

“…Shah et al [32] COVID-19 detection from X-ray images Help in diagnosing potential suspects as early as possible to mitigate the deadly disease Ashraf et al [33] Predict the severity of disease or chances of death Significant contribution in separating vulnerable groups for ample care Prakash et al [34] Impact analysis of various policies employed to control the disease Guidance for effective strategies that can help control the spread Ullah et al [35] Classification of patients with and without COVID-19 Lowering the healthcare burden of patient diagnosis Rathod et al [36] Effective crisis preparedness and management along with authorities' responses and mitigation strategies. Assistance in healthcare workers' burden analysis Rathod et al [37] Detection of abnormal data for effective analysis Resource planning and accurate diagnosis ML Rashed et al [38] Provides public awareness about the morbidity risk of COVID-19 Consistent and reliable forecasting patterns of the spread/decay phases of COVID-19 Hu et al [39] Feasible analysis model for the treatment and diagnosis of COVID-19 Effective identifications of key symptoms and medicines for different syndromes Singh et al [40] Reduce the high false-negative results of the RT-PCR Effectively handles the sensitivity issue that is associated with RT-PCR Peddinti et al [41] Detection of COVID-19 cases in public places Helps officials in the accurate and faster diagnosis of the virus Saverino et al [42] Changes implementation in rehabilitation services Staff satisfaction and stress reduction during pandemic times Lella et al [43] Respiratory sound classification for potential patient identification Classification of asthma sounds, COVID-19 sounds, and regular healthy sounds Malla et al [44] Real-time sentiment analysis of COVID-19 tweets Tweets prediction related to similar types of infectious diseases in the future Ibrahim et al [45] Accurately diagnosing COVID-19 patients and analyzing severity level Detecting COVID-19 patients and classifying the severity degree from chest CT slices Roland et al [46] Blood-test-based identification of patients with COVID-19 and estimate the mortality risk Automatic scanning of COVID-19 in a cost-effective way without any additional efforts Gros et al [47] Accurate estimates of the cumulative medical load of COVID-19 outbreaks Understanding the outbreak dynamics and predicting future cases and fatalities Hack et al [48] Promising treatments, including the virus' protein structure and attack mechanisms analysis, and resource planning Accelerate the science needed to develop treatments and strategies to combat COVID-19 West et al [49] COVID-19 spread analysis among different populations and effective therapeutic response Virus transmissi...…”