The AI ​​tool interprets ECG images with precision at the pixel level

Electrocardiogram (ECG) is one of the most essential tools in modern medicine, used to detect heart problems ranging from arrhythmias to structural anomalies. In the United States only, millions of ECGs are carried out each year, whether in emergency rooms or routine doctor visits. As artificial intelligence systems (AI) (AI) become more advanced, they are increasingly used to analyze ECGs, sometimes even detecting the conditions that doctors may miss.

The problem with this is that doctors must understand why an AI system makes a certain diagnosis. Although the ECG analysis fueled by AI can achieve great precision, it often works as a “black box”, giving results without explaining its reasoning.

Without clear explanations, doctors hesitate to trust these tools. To fill this gap, the researchers of the Technion work to make AI more interpretable, which gives it the capacity to explain its conclusions in a way that aligns with medical knowledge.

Make AI talk about the doctor’s language

For AI to be useful in clinical environment, it should highlight the same characteristics of the ECG on doctors when diagnosing heart disease. This is difficult because even in cardiologists, there is not always a complete agreement on the most important ECG markers.

Despite this, the researchers have developed several interpretation techniques to help AI explain its decisions. But these techniques sometimes highlight the large regions of the ECG, without identifying the exact marker, leading to potential erroneous interpretations. They also sometimes highlight non-relevant parts of the image, such as background, rather than the real ECG signal.

The next step: AI for real world ECGs

The most recent AI models are based on high -quality digitized ECG images. But in the real world, doctors do not always have access to perfect analyzes. They often rely on paper prints from ECG machines, which they could photograph with a smartphone to share with colleagues or add to a patient’s files. These photographed images can be tilted, crumpled or shaded, which makes IA analysis much more difficult.

To solve this problem, Dr. Vadim Glour, a former doctorate. The student of the Biomedical Engineering Laboratory of Professor Yael Yaniv at Technion, in collaboration with the Schuster laboratory of the IT faculty of Henry and Marilyn Taub, has developed a new AI interpretation tool designed specifically for ECG images photographed.

This article was published in NPJ Digital Medicine. Using an advanced mathematical technique (based on the Jacobian matrix), this method offers precision at the level of the pixel, which means that it can highlight even the smallest details of an ECG. Unlike previous models, it is not distracted by the background and can even explain why certain conditions do not appear in a given ECG.

A more transparent future for medical AI

While AI continues to play a more important role in health care, making it explainable and trustworthy is just as important as making it precise. By developing methods that allow AI to communicate its results in a way that aligns with medical expertise, researchers help to open the way to smarter, more reliable and more widely accepted AI tools in cardiology.

With this progress, doctors could soon have AI assistants who not only detect heart problems, but also clearly explain their reasoning, leading to better, faster and more informed patients.