Reliably predicting progression of Covid-19

Inselspital, Bern University Hospital and the University of Bern are currently launching the world’s first multicenter, international study on AI-assisted prediction of severe progressions of Covid-19. The research uses artificial intelligence to evaluate extensive clinical, image-morphological and laboratory data. The study, which is funded by the Swiss National Science Foundation, aims to provide reliable predictions as to whether a specific case would lead to a severe progression of Covid-19.

Since the first appearance of Covid-19, our knowledge regarding SARS-CoV-2 and the different manifestations of the disease caused by the virus has increased. To date, there is a lack of clear understanding of why some patients only experience mild symptoms, while others have to be treated for severe (acute or chronic), and in the worst cases lethal, progressions. These questions are now being addressed in a project (NRP-78 «COVID») funded by the Swiss National Science Foundation.

New research approach

The international team from Inselspital Bern, Bern University Hospital, the University of Bern, the University of Parma (IT) and Yale University (USA) is working on a system based on artificial intelligence (AI).

The new AI system processes information from chest CT and conventional X-ray images, laboratory parameters and clinical data. The system should predict the seven-day progression in the acute phase on the basis of a patient’s current condition. Furthermore, it ought to enable statements on the medium- and long-term progression and chronicity of the disease.

The AI-based approach with complex input data aims to better understand the progression of a disease. The goal of the project is a faster and more accurate prognostic grouping of COVID-19 patients based on the proposed AI system. This would enable faster and more appropriate treatment for Covid-19 patients.

Combining “multi-omics” and AI

A new feature of the project presented here is the inclusion of extensive baseline biological data, the so-called “omes” 1). According to the research hypothesis, the analysis of the very extensive data by means of AI enables the detection of previously unknown correlations between physiological, image-morphological and pathological facts. In this way, the researchers hope to identify new biomarkers that are crucial for assessing the progression of Covid-19.

Bioinformatician and AI expert Prof. Mauricio Reyes clarifies: “We need to find ways to ensure that our deep learning systems function independently of specific hospital centers and device types or methods of analysis. With our project, we are pushing forward in two directions: We are using data from three of the world’s leading Covid research centers, thus increasing the amount of data. Moreover, we are integrating CT and classic X-ray images with a multi-omics approach, thereby broadening the underlying information from a technical point of view. In this way, we hope to arrive at new, more reliable and faster predictions of a Covid progression.”

Large data sets provide sufficient learning environment for AI algorithms

One of the central problems with using AI in medicine is limited access to data. Algorithms rely on large sample quantities to produce qualitatively sound results. Furthermore, different systems must be able to be included in the learning process, otherwise only limited statements can ultimately be made about a certain type of device or a certain location. For this reason, Inselspital, Bern University Hospital and the University of Bern are joining forces with the Universities of Parma (IT) and Yale (USA).

There have only been a few studies with an AI multi-omics approach to COVID-19 thus far, and compared to these studies, the research team will use a much more extensive set of data. Also new is the approach of combining computed tomographic and conventional X-ray images to better assess the imaged progression of the disease. Moreover, this study will link several centers for the first time (multicenter study). This is expected to render the results sounder and generally transferable.

Next steps

An intensive data collection process is currently under way. Prof. Dr. Alexander Pöllinger, head of the AI multi-omics project, explains: “Our AI multi-omics project addresses the problem of the limited amount of available information in order to study COVID-19 progression with a proper approach. To this end, we are basically integrating all available Information sources from three major international hospital centers.”

Bernese research and CAIM

Other Bernese researchers have used AI to develop new Covid-19 diagnostic methods (i.e., AI X-ray image analysis detects Covid-19 more reliably). Efforts are under way worldwide to network major data centers more efficiently and to create virtual platforms. For years, the medical center in Bern has been working on consistently promoting and intensively networking AI. The preliminary highlight is the founding of the Center for Artificial Intelligence in Medicine CAIM on March 19, 2021. Please visit the website (registration is required). 

 

1)  Overview multi-omics methodology: and explanation in general: Multiomics

 

Experts:

  • Prof. Dr. med. Alexander Pöllinger, Senior Consultant, University Institute of Diagnostic, Interventional and Pediatric Radiology, Inselspital, Bern University Hospital
  • Prof. Dr. Mauricio Reyes, ARTORG Center for Biomedical Engineering, University of Bern

 

Contact: 

AI pattern recognition, algorithm learning process

Multi-omics approach to the study. Complex data, consisting of CT, X-ray data, clinical and laboratory data, serve the AI algorithm as a basis for the prognosis of the acute (7-day) and chronic course

Segmentation of lung changes in COVID-19 patients. The AI system is "trained" with this data set.

Prof. Mauricio Reyes and Prof. Alexander Pöllinger analyze CT images from a lung examination.

« back to list