A revolutionary AI innovation called HeartBEiT, developed by researchers at Mount Sinai, is significantly advancing the accuracy and detail of electrocardiogram (ECG) diagnoses, even for rare conditions with limited data. By interpreting ECGs as language, this cutting-edge model outperforms traditional convolutional neural networks (CNNs) by pinpointing specific areas of the ECG that are responsible for heart-related conditions.
The researchers at Mount Sinai have introduced an innovative artificial intelligence (AI) model for the analysis of electrocardiograms (ECGs) that treats them as a form of language. This novel approach has the potential to improve the accuracy and efficacy of ECG-based diagnoses, particularly for cardiac conditions where there is a scarcity of training data.
In a recent study published in npj Digital Medicine on June 6, the team unveiled their deep learning model called HeartBEiT, which serves as a foundation for creating specialized diagnostic models. Comparative tests demonstrated that models built using HeartBEiT outperformed established methods for ECG analysis.
Lead author Dr. Akhil Vaid, MD, who is an Instructor of Data-Driven and Digital Medicine (D3M) at the Icahn School of Medicine at Mount Sinai, stated, “Our model consistently outperformed convolutional neural networks [CNNs], which are commonly used machine learning algorithms for computer vision tasks. These CNNs are often pretrained on publicly available images of real-world objects.” Dr. Vaid further explained that HeartBEiT’s specialization in ECGs enables it to achieve comparable or superior performance using a fraction of the data, making ECG-based diagnosis more feasible, particularly for rare conditions that affect a smaller number of patients and consequently have limited available data.
Electrocardiograms are widely used due to their affordability, non-invasive nature, and applicability to various cardiac diseases, with over 100 million ECGs performed annually in the United States alone. However, the usefulness of ECGs is limited because physicians often struggle to consistently identify disease patterns visually, especially for conditions lacking established diagnostic criteria or where patterns are subtle or chaotic. Fortunately, artificial intelligence is now revolutionizing this field, with most of the current research focused on CNNs.
Mount Sinai is taking a bold step forward by leveraging the growing interest in generative AI systems, such as ChatGPT, which rely on transformer-based deep learning models trained on vast text datasets to generate human-like responses. In this case, researchers are using a related image-generating model to represent discrete portions of the ECG, enabling its analysis as a language.
Dr. Vaid explained, “These representations may be considered individual words, and the whole ECG a single document. HeartBEiT understands the relationships between these representations and uses this understanding to perform downstream diagnostic tasks more effectively.” The study tested the model on three tasks: determining if a patient is experiencing a heart attack, identifying a genetic disorder called hypertrophic cardiomyopathy, and assessing heart function. In each instance, HeartBEiT outperformed all other baselines.
To train HeartBEiT, researchers used 8.5 million ECGs from 2.1 million patients collected over four decades from four hospitals within the Mount Sinai Health System. Subsequently, they evaluated its performance against standard CNN architectures in the three areas of cardiac diagnosis. The study demonstrated that HeartBEiT exhibited significantly higher performance with smaller sample sizes and provided better “explainability.” Senior author Dr. Girish Nadkarni, MD, MPH, highlighted, “Neural networks are considered black boxes, but our model was much more specific in highlighting the region of the ECG responsible for a diagnosis, such as a heart attack, which helps clinicians better understand the underlying pathology. By comparison, the CNN explanations were vague even when they correctly identified a diagnosis.”
Through their sophisticated modeling architecture, the Mount Sinai team has greatly improved the interaction between physicians and ECGs, presenting new possibilities for analysis. Dr. Nadkarni emphasized, “We want to be clear that artificial intelligence is by no means replacing diagnosis by professionals from ECGs but rather augmenting the ability of that medium in an exciting and compelling new way to detect heart problems and monitor the heart’s health.”
The research paper, titled “A foundational vision transformer improves diagnostic performance for electrocardiograms,” was funded by the National Heart, Lung, and Blood Institute of the NIH (grant number R01HL155915) and the National Center for Advancing Translational Sciences of the NIH (grant number UL1TR004419).
Table of Contents
Frequently Asked Questions (FAQs) about AI-powered ECG analysis
What is HeartBEiT?
HeartBEiT is an AI model developed by researchers at Mount Sinai. It is designed to analyze electrocardiograms (ECGs) by interpreting them as language, enhancing the accuracy and detail of ECG-based diagnoses.
How does HeartBEiT improve ECG diagnoses?
HeartBEiT outperforms traditional CNNs (convolutional neural networks) by highlighting specific areas of the ECG that are responsible for heart conditions. It provides a more precise analysis, especially for rare conditions with limited data, and enhances the effectiveness of ECG-related diagnoses.
How was HeartBEiT trained?
Researchers pretrained HeartBEiT on 8.5 million ECGs from 2.1 million patients collected over four decades from four hospitals within the Mount Sinai Health System. This extensive training dataset allows the model to learn patterns and relationships in ECGs, enabling accurate diagnostic performance.
What diagnostic tasks can HeartBEiT perform?
HeartBEiT has been tested on three diagnostic tasks: determining if a patient is having a heart attack, identifying hypertrophic cardiomyopathy (a genetic disorder), and assessing heart function. In all these cases, HeartBEiT performed better than other tested baselines.
How does HeartBEiT contribute to ECG analysis?
HeartBEiT enhances the way physicians interact with ECGs by representing them as language. It improves the explainability of diagnostic results, highlighting the specific region of the ECG responsible for a diagnosis. This can aid clinicians in understanding the underlying pathology of heart conditions.
Is artificial intelligence replacing human diagnosis in ECG analysis?
No, the purpose of HeartBEiT and similar AI models is not to replace human diagnosis but to augment the abilities of medical professionals. By providing accurate and detailed analysis, these AI models assist doctors in detecting heart problems and monitoring the heart’s health in a more effective and compelling manner.
More about AI-powered ECG analysis
- Mount Sinai Health System: Official Website
- npj Digital Medicine: Journal Article
- National Heart, Lung, and Blood Institute (NIH): Official Website
- National Center for Advancing Translational Sciences (NIH): Official Website
