Using machine learning in the fight against Covid-19

Category: Englisch

Dr. Anna Laura Herzog from the University Hospital Würzburg and Prof. Dr. Holger v. Jouanne-Diedrich from TH Aschaffenburg are the main drivers behind a recent study on the use of machine learning to predict the course of partial aspects of a Covid 19 infection.

Illustration from the paper on determining the BMI threshold using the OneR machine learning package

An algorithm developed at TH Aschaffenburg offers valuable support in the treatment of corona patients


As part of the lecture series on the topic of "Artificial Intelligence and Digitalisation in Healthcare", Prof. Dr Holger v. Jouanne-Diedrich from Aschaffenburg University of Applied Sciences and Dr Anna Laura Herzog from the University Hospital of Würzburg (UKW) presented various machine learning methods that make it easier to make the right clinical decisions when treating Covid 19 patients.

Covid 19 is a multisystem disease whose severity and course depends on the type and number of organ systems involved. Various risk factors such as obesity, high blood pressure and advanced age worsen the course of the disease, and if the heart and kidneys are also involved, the risk of dying from a Covid 19 infection increases dramatically.

Based on evidence that the SARS-CoV2 virus is often associated with kidney failure, university professor von Jouanne-Diedrich and head of the UKW Transplant Centre Dr Herzog investigated whether the presence of proteinuria (excessive excretion of protein in the urine) could be used to predict kidney failure, the development of chronic kidney disease and mortality in critically ill COVID 19 patients. To do this, they used machine learning (ML) methods, some of which were developed at TH Aschaffenburg. 

Algorithm publicly available free of charge
The OneR package developed by Jouanne-Diedrich makes it easy to find influencing factors and limit values (cut-off points). "I am proud that the OneR package can make a contribution in the fight against the pandemic," says the delighted professor, who teaches and conducts research in the field of artificial intelligence at TH Aschaffenburg and helped conceive, establish and design the new Medical Engineering and Data Science degree programme. "I therefore made the package available to the interested public free of charge some time ago."
The special feature of the newly developed procedure is that the results are presented in the form of easily understandable rules. This often makes it superior to more complicated methods, such as "neural networks" (so-called "deep learning"), which are difficult to comprehend. It is not only in the medical context that interpreting the results easily is of great importance. 

Protein loss as an important predictor variable
The study investigated whether kidney failure can be predicted in the case of severe Covid 19 infection and whether there are blood values from routine treatment that can predict the course. Among other things, the ML algorithm was able to identify protein loss, i.e. kidney involvement, as a valuable variable for predicting progression and thus predicting whether longer-term chronic kidney disease is to be expected.
All results of the joint research were published in the scientific journal PLOS One and can be accessed freely: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0251932