Researchers at the Technion laid down the principles for the correct development of artificial intelligence-based tools for the world of medicine and showed how to use them to develop useful systems in the world of cardiology
In recent years, meteoric progress has been made in the fields of deep learning, but currently there are almost no medical products on the shelf that use such technology; Thus, doctors continue to perform their work as they were used to in previous decades.
To find a solution to the problem, the group of Prof. Yael Yaniv from the Faculty of Biomedical Engineering joined the research groups of Professors Alex Bronstein and Asaf Shuster from the Taub Faculty of Computer Science. Now, under their joint guidance, a study by doctoral students Yonatan Elul and Aviv Rosenberg has been published in the journal of the American Academy of Sciences - PNAS.
In an article, the researchers demonstrate an artificial intelligence-based system that automatically detects diseases based on hundreds of ECG charts - the most common technology today for diagnosing heart problems.
The new system automatically analyzes the ECG records using layered neural networks - the most prominent tool in the field of computational learning today. These networks learn different patterns by training on multiple examples, and the system developed by the researchers was trained on more than 1.5 million ECG segments sampled from hundreds of patients from hospitals in different countries.
The EKG test, which was developed more than a century ago, quickly provides important data on the state of the heart, quickly and without the need for invasive surgery. The problem is that the reading of the registry is currently done by a human cardiologist, and thus subjective elements permeate the decoding of the registry. That is why many research groups in the world are working to develop systems that will perform efficient and accurate automatic decoding. More so, due to its capabilities to analyze multiple data The system succeeds in identifying disease states that human cardiologists, however experienced, would not be able to identify.
The system developed by the researchers was built according to requirements defined by expert cardiologists and it produces an output that includes the degree of certainty of the results, marking of suspicious areas on the EKG wave and warnings of inconclusive results and increased risks of pathology not observed in the EKG signal itself. The system demonstrates sufficient sensitivity in alerting subjects at risk of arrhythmias even if the arrhythmia is not in the recording, yet it almost never provides false alerts. Furthermore, the new system explains its decisions in terms accepted in the world of cardiology.
The researchers hope that this system can be used for cross-sectional scans in the population for the early identification of people at risk of suffering from arrhythmias. Without such early diagnosis, these people are at increased risk of heart attacks and strokes.
The research was led by Prof. Yael Yaniv, head of the laboratory for bioenergetic and bioelectrical systems at the Faculty of Medical Engineering; Prof. Alex Bronstein, head of the VISTA laboratory at the Taub Faculty of Computer Sciences; Prof. Assaf Shuster, the Laboratory for Learning in Giant Properties at the Taub Faculty of Computer Science and co-head of the Center for Computational Learning; Yonatan Elul, a doctoral student in the laboratories of Prof. Bronstein, Prof. Yaniv and Prof. Schuster, who completed a bachelor's degree in biomedical engineering and a master's degree in the Faculty of Computer Science at the Technion; and Aviv Rosenberg, a doctoral student in the laboratory of Prof. Bronstein and Prof. Yaniv, who completed a bachelor's degree at the Technion in the Viterbi Faculty of Electrical and Computer Engineering and a master's degree in the Faculty of Biomedical Engineering.
The research was supported by the Ministry of Science and Technology and the Cyber Foundation. The study took place within MLIS - Center for Computational Learning and Intelligent Systems, which integrates all artificial intelligence activity at the Technion. In recent years, the Technion has been ranked at the top of the world in research and development in AI, and the prestigious international ranking CSRankings places it in first place in Europe (and of course in Israel), and in 15th place in the world in this field. In the subfield of machine learning, the Technion ranks even higher, in 11th place in the world. This is based on 2021-2016 data. Today, 46 researchers at the Technion are engaged in the core areas of AI and more than 100 researchers work in fields related to AI. MLIS is headed by Prof. Shai Menor from the Viterbi Faculty of Electrical and Computer Engineering and Prof. Assaf Schuster from the Taub Faculty of Computer Science.
For an article in the journal PNAS
More of the topic in Hayadan:
One response
Finally!