At the AI Cures Conference, Professor Yonina Eldar will present exciting updates on her and her team’s research on innovative methods for point-of-care image analysis for COVID-19. Their project’s goal was to develop and implement image analysis techniques using artificial intelligence that will help with identification, triage and diagnosis of COVID-19 patients or suspected carriers.
Using a one-of-a-kind data set of thousands of X-ray images of COVID-19 patients in Israel, the team of data science experts and physicians are learning features that help identify COVID-19 symptoms and help in determining the level of disease. With their algorithm, they are already able to obtain around 90% accuracy on patients admitted to hospitals, which far surpasses the detection rate of the standard PCR tests that is around 70% for virus carriers.
Speed is another benefit of this technique versus PCR testing. The results can be readily obtained on-site at the hospital, from the X-ray images, whereas PCR tests require several hours or days to process, usually in remote laboratories.
Hospitals already have X-ray machines. They are easy to operate and easy to sanitize. The part that is difficult is reading the films and differentiating COVID-19 from among a number of other conditions. It is here that the AI, with the benefit of learning from thousands of scans, has the edge over a human operator. These methods will also be used to monitor disease progression and prognosis as well as to monitor patients post-disease.
A follow-on project focuses on creating AI to interpret ultrasound scans to diagnose COVID-19. Ultrasound machines depend on a skilled operator, however they have benefits that make them ideal for use in the community versus an X-ray. An ultrasound machine is portable and the probe is easily sanitized making it an ideal method to use outside a hospital. If the algorithm is valid, one or two trained operators can have a sizeable impact on community testing that just is not as easily done in a hospital.
Ultrasound is not yet a common tool for diagnosis in Israel, but the team’s goal is to rate the severity of disease based on ultrasound alone. By using a combination of computer vision and artificial intelligence techniques, they can identify irregular pleural lines and consolidations as well as other indicators for respiratory distress. They are getting very promising results and hope that this will pave the way to more pervasive use of ultrasound monitoring for lung patients.
Not content with creating AI models for currently available scanning technology, Eldar and team are also exploring simple methods for COVID-19 detection such as voice and heart-rate via radar signals. This research is in its nascent stages, yet has the potential for a huge impact.
COVID-19 is the focus of their research, however, the hope is that artificial intelligence can be used with imaging more generally within hospitals and communities. This may lead to more effective and efficient diagnosis and prediction of disease severity for pulmonary patients.