Technion researchers have discovered a viable way in which to reliably use Artificial Intelligence (AI) in medicine and demonstrated the use of practical systems for cardiology in an article published in Proceedings of the National Academy of Sciences of the United States of America (PNAS).
Even as AI has developed greatly over the past decades, the use of this technology in medical products is still scarce, and the methods currently employed by doctors are based on older technology.
In the article, the research team demonstrated the use of the new technology to identify diseases based on hundreds of electrocardiograms (ECG), which are currently the most common technology used in cardio-medicine.
The new systems analyzes the ECGs using an augmented neural network, which has been taught various patterns based on a system of more than 1.5 million ECG tests from hundreds of patients in hospitals worldwide.
The ECG is a quick and non-invasive test that provides information on the workings of the heart. The disadvantage of the test is that the cardiologists who read the printouts are susceptible to mistakes in their interpretations, either because they are subjective or because they cannot see what they are looking for with enough precision.
The new systems are more accurate and can detect pathological conditions that human cardiologists are physically unable to see.
The researchers worked closely with cardiologists and created the system according to their requirements. The output includes an uncertainty estimation of results, alerts regarding inconclusive results and increased risk of pathology that the ECG signal does not observe itself.
The system demonstrated enough sensitivity and precision that it can alert for risk of arrhythmia even when it is not demonstrated in the ECG. Without this early diagnosis, people run a higher risk of heart-attacks and strokes.
Moreover, the AI uses official cardiology terminology to explain its decisions.
The study was headed by Prof. Yael Yaniv, director of the Bio-electric and Bio-energetic Systems Laboratory at the Faculty of Biomedical Engineering at the Technion; Prof. Alex Bronstein, director of the VISTA Laboratory at the Taub Faculty of Computer Science; Prof. Assaf Schuster of the Learning at Scale Laboratory (MLL) at the Taub Faculty of Computer Science and co-director of the MLIS Center (Machine Learning & Intelligent Systems); Yonatan Elul, a doctoral student in the laboratories of Professors Bronstein, Yaniv, and Schuster and Aviv Rosenberg, a doctoral student in the laboratory of Professors Bronstein and Yaniv.
The project was sponsored by the Ministry of Science and Technology and the Technion Hiroshi Fujiwara Cyber Security Research Center and the Israel Cyber Directorate.