Radiology battling COVID-19

Specialists working for medical projects at Mindy Support share their views on the potential of AI to help combat COVID-19.

Anastasia Budkina, PM for Medical Data Annotation Projects

Every hour is critical when it comes to detecting coronavirus. Medical staff throughout the world are overwhelmed with the flow of patients with lung diseases and their complications, and we believe that during these hard times, AI can be a turning point in helping to stop the spread of the virus. Exhausted doctors have to make quick, precise diagnoses during day-long shifts, and automated software can be the tool to make their work more efficient.

Solutions to detect COVID-19 in chest CT scans, differentiate it from other diseases, and determine how the patient’s condition will change in the upcomings days – all these features need to be addressed, and as soon as possible. Many European hospitals have started to collaborate with medical startups and AI companies to bring these ideas to life. We truly believe that such efforts will help us conquer the pandemic.

Iryna Hlado, Radiologist

In the context of the pandemic, I have encountered projects such as COVID-19 Open Research Dataset and the COVID-Net open-access neural network. The latter also includes working on pulmonary tissue pathologies using CT computed tomography, in attempts to develop a system identification of COVID-19. In China, Alibaba created an AI-based diagnostic system similar to a previous one conceived by Israeli startup RADlogics.

There are patterns that can help us to identify COVID-19; the CORADS Classification describes the six stages of pathology, plus typical and atypical features. Where AI helps is with the speed of such identification. Unlike human resources, AI is not restricted by fatigue, stress or the diversion of attention that occurs from seeing a large number of patients. Its work is predictable, well-programmed and efficacious. Humans, of course, are creative beings blessed with abstract thinking, able to identify a disease that is atypical.

A quote from the World Economic Forum is worth considering: “AI can help with the COVID-19 crisis – but the right human input is a key.” In other words, we should work together. This is the key to success.

In my opinion, the main barrier in medical research is that the same pattern can be characteristic of many diseases. Thus, AI has identified a pathological pattern that suits a number of diagnostic diseases. Actually, the medical contribution is necessary in the form of a conclusion, which indicates a single correct diagnosis. Anamnesis – questions of whether the patient has traveled abroad, whether they have been in contact with sick people, etc. – will help, as will laboratory diagnostics.

Roman Marin, Radiologist

Taking into account the huge volumes of data obtained from various medical organizations, we can try to analyze the possibilities, advantages and shortcomings of AI for diagnosing COVID-19 through CT scans. Many people believe that AI can already predict the development of acute respiratory distress syndrome with a very high degree of accuracy (up to 80%). Undoubtedly these algorithms are based on a combination of changes in the lung tissue, with clinical trials where high levels of alanine transaminase and myalgia were observed along with increased hemoglobin levels. Fever and lymphopenia have a less apparent diagnostic profile.

Any pandemic requires an all-hands-on-deck effort in terms of labor and resources to find proper solutions. There has been lots of scientific information gathered since December 2019 revealing how quickly the virus spreads and the fatality rate, the latter of which remains unclear. In any case, greater volumes of information should hasten the development of novel technologies. Today, pretty much all AI projects are done in real-time i.e. assessing the effectiveness, analyzing mistakes, upgrading the algorithms – all of this done in a single day.

Any scientific analysis is a multifaceted and complex process, but even more so today with COVID-19 proliferating and given the complexities of the virus and its two-step infection process. Supercomputers such as those built at NASA and the IBM Summit supercomputer are being utilized to work on solutions. The use of AI to solve such problems is a testament to its enormous scientific potential. Problems concerning the development and application of AI algorithms are mostly due to data protection and privacy, fairness and transparency.