New AI Algorithm Detects Heart Attacks in 1 Second: Promising Breakthrough in Cardiac Diagnostics
A team of researchers at ITMO’s Research Center Strong AI in Industry has developed an innovative algorithm that can detect symptoms of a heart attack in just one second. Trained on over 20,000 electrocardiograms (ECGs), the algorithm boasts an accuracy rate of 85% and can immediately pinpoint sections of an ECG recording that may be associated with a heart attack.
Cardiovascular diseases are a leading cause of death worldwide, claiming approximately 17.9 million lives annually, with more than half of these deaths attributed to heart attacks. A heart attack occurs when there is a disruption in blood supply to the cardiac muscle. Prompt medical attention is crucial in these cases, making it essential to call for an ambulance at the first signs of a heart attack.
Currently, heart attacks can be diagnosed using an ECG, which detects bloodflow-related changes in cardiac function. However, additional tests at a laboratory are often required to confirm the diagnosis, making it time-consuming during emergencies when immediate action is necessary.
To address this issue, numerous AI methods are being developed to assist in ECG-based diagnostics. For example, a group of researchers from Stanford University and the University of California previously created a deep neural network capable of automatically detecting and classifying arrhythmias. Another example is an algorithm developed by researchers from the South Korean Sejong Medical Research Institute in 2020, which can detect a heart attack using a six-lead ECG.
At ITMO, the research team focused on developing their own algorithm to aid clinicians in diagnosing heart attacks based on ECG data. They utilized the classification of QRS complexes, which register a single heart cycle, as the basis for their algorithm. Traditionally, QRS complex classification only predicted specific properties, but the updated model now identifies patterns characteristic of a heart attack in an entire ECG recording.
To achieve this, the researchers wrote software in PyTorch, a machine learning framework, and employed a Siamese neural network. The algorithm was then trained on a dataset consisting of ECGs from both healthy patients and those with heart attacks. The model demonstrated accurate results in 85% of cases and required just one second to make a diagnosis based on an ECG sample.
Furthermore, the researchers integrated the Grad-CAM tool into the model, which interprets and visualizes the results provided by the algorithm. This addition enables the algorithm to produce an ECG graph highlighting sections that contain patterns associated with a heart attack.
According to Natalia Gusarova, the head of the project, the algorithm can make a decision based on ECG data from as few as five heartbeats. This means that a five-second recording is now sufficient for diagnosis compared to the standard 20-second recording.
Vladimir Shilonosov, one of the algorithm’s developers, explains that another advantage of the new solution is its use of FewShot technology. This allows the model to be further trained on specific data, such as ECGs from patients at a particular hospital, thereby increasing its accuracy. The model can be tailored for use in different medical institutions, taking into account variations in data parameters, ECG devices, and patient groups.
The algorithm is intended to serve as a support system for faster and more accurate heart attack diagnostics, rather than replacing healthcare professionals. Clinical trials are already planned at the Almazov National Medical Research Center to evaluate the algorithm’s effectiveness.
Looking ahead, the research team aims to train the model to detect other cardiovascular diseases, such as stenocardia and coronary heart disease.
Vladimir Shilonosov emphasizes the transparency of their algorithm compared to other AI-based decision support systems. Their model incorporates explainable AI, providing clinicians with a step-by-step analysis and enhancing trust in the results. The algorithm generates a highlighted ECG graph as the basis for its decision-making process.
The groundbreaking AI algorithm developed by ITMO’s Research Center Strong AI in Industry offers a significant advancement in diagnosing heart attacks promptly and accurately. With its ability to detect symptoms in just one second, the algorithm has the potential to revolutionize cardiac diagnostics and potentially save countless lives.
