The Pulse of Progress

Heart problems pose a significant burden on individuals and societies worldwide, constituting a major global health challenge. According to the World Health Organization, cardiovascular diseases are the leading cause of death globally, accounting for approximately 31% of all global deaths. This fact underscores the pervasive impact of heart-related issues, affecting millions of lives and straining healthcare systems.

Early detection of heart problems is crucial, as timely intervention often leads to better long-term outcomes. Regular screenings and diagnostic assessments can identify risk factors and underlying conditions before they escalate into more severe health issues. However, a significant challenge lies in the fact that many heart conditions initially manifest without noticeable symptoms, making it difficult for individuals to recognize potential problems and seek medical attention proactively.

Furthermore, the complexity and expense associated with traditional imaging techniques used to assess heart function create additional barriers to their widespread administration, especially in cases where no apparent problem is suspected. These factors hinder the ability of healthcare professionals to detect subtle abnormalities and initiate early treatment, as routine screenings may not be feasible on a large scale due to resource constraints.

Researchers at the Icahn School of Medicine at Mount Sinai suspected that simpler, less expensive testing procedures might be able to detect heart conditions with the same level of accuracy as traditional methods if they are paired with a deep-learning algorithm that can help to interpret the data. They conducted a study in which electrocardiogram (ECG) measurements — which can easily be collected in a wide range of clinical settings — were interpreted by a deep-learning model to assess the health of the heart’s right ventricle.

Initially, the team chose to look at critical factors, like the size of the right ventricle, and its ability to pump blood normally. These parameters are typically challenging to assess, so to test their theory, the researchers trained a machine learning model on a large dataset consisting of 12-lead ECGs and cardiac magnetic resonance imaging (MRI) measurements. The MRI measurements served as the ground truth data to help the model learn to recognize the correspondence between ECG signals and abnormalities with the right ventricle.

Specifically, the model was trained to numerically predict both the right ventricular ejection fraction and end‐diastolic volume. A four month study was conducted to assess the accuracy of the system, and it was discovered that the model performed well in estimating these metrics. However, the team notes that their work is in the early stages and cannot yet replace traditional, advanced diagnostic tests. Further research will be needed to assess the tool’s safety and accuracy before it can be used in real-world scenarios.

Looking ahead, the team plans to perform additional validations of their system in diverse populations to ensure that it is generally applicable. They also intend to assess how well their model can detect conditions like pulmonary hypertension, congenital heart disease, and various forms of cardiomyopathy. A simple, inexpensive way to screen for these conditions could be a key component in reducing the burden of heart-related medical conditions in the future.