MIT Team Develops AI Tool to Forecast Heart Failure Progression
Researchers at MIT, Mass General Brigham, and Harvard Medical School have developed a deep learning model called PULSE-HF to predict heart function deterioration in heart failure patients. The AI tool analyzes electrocardiograms to forecast declines in left ventricular ejection fraction, potentially helping clinicians prioritize care. Early results demonstrate high predictive accuracy across multiple hospital datasets.
A team of researchers from MIT, Mass General Brigham, and Harvard Medical School have created an artificial intelligence tool aimed at transforming the management of heart failure, one of the persistent challenges in clinical medicine. The model, known as PULSE-HF (Predict changes in left ventricULar Systolic function from ECGs of patients who have Heart Failure), leverages deep learning—a form of machine learning that uses neural networks modeled after the human brain—to analyze electrocardiograms (ECGs) and forecast which patients are at risk of worsening heart function within a year.
Heart failure, characterized by a weakened or damaged heart muscle, remains incurable and leads to significant health burdens, including high rates of hospitalization and mortality. According to Teya Bergamaschi, a PhD student at MIT and co-first author of the study, roughly half of all diagnosed patients die within five years of diagnosis. Given finite healthcare resources, timely identification of patients most likely to deteriorate is critical.
The research, recently published in Lancet eClinical Medicine, details how PULSE-HF was trained and retrospectively tested on data from three major sources: Massachusetts General Hospital, Brigham and Women's Hospital, and the public MIMIC-IV dataset. The model’s primary target is the prediction of changes in the left ventricular ejection fraction (LVEF), a key measurement indicating how much blood the heart's left ventricle pumps out with each contraction. An LVEF below 40 percent is a marker of severe heart failure.
"The model takes an ECG and outputs a prediction of whether or not there will be an ejection fraction within the next year that falls below 40 percent," said Tiffany Yau, MIT PhD student and co-first author. This advance enables clinicians to focus their attention on higher-risk patients, potentially reducing unnecessary hospital visits and allowing efficient use of resources, especially in lower-resource or rural settings.
Unlike past AI methods that simply detect existing heart damage, PULSE-HF is designed for forecasting, marking a novel step in the application of predictive analytics in healthcare. The deep learning system achieved strong results, with area under the receiver operating characteristic curve (AUROC) scores between 0.87 and 0.91 across all three patient cohorts. AUROC values closer to 1 indicate excellent predictive performance.
Notably, the team developed versions of the model compatible with both traditional 12-lead ECGs and more limited single-lead ECGs. Despite the reduced input data, the single-lead version performed comparably to the full 12-lead model, broadening potential future applications in primary care or settings without specialized staff.
Development was not without obstacles. Gathering, processing, and labelling clinical ECG and echocardiogram data proved challenging due to irregular file formats and data quality issues—challenges typical in real-world healthcare AI development. Bergamaschi noted that the team invested significant time in cleaning data and addressing signal artifacts.
The next phase involves prospective trials—testing PULSE-HF in real patient care settings where true future outcomes are unknown at the time of prediction. If successful, such AI-driven tools may play an increasing role in personalizing treatment plans for patients with chronic conditions like heart failure.
For now, the work demonstrates both the complexity and the promise of applying deep learning to healthcare prediction tasks, potentially helping clinicians more effectively manage a widespread, life-threatening disease.
Related Posts
Multi-Agent Deep Reinforcement Learning Audits Silver Futures Markets
A new technical implementation introduces a multi-agent audit engine leveraging double deep Q-networks to monitor strategic behaviour in the Silver futures market. The system uses advanced deep reinforcement learning and recent research to distinguish between competitive and potentially cooperative trading agent outcomes. Results indicate that during a recent sample window, market behaviors remained within normal competitive ranges rather than exhibiting signs of tacit coordination.
AI Model Identifies Biological Markers for Human Longevity
A recent study demonstrates how artificial intelligence can detect biological indicators that may predict individual lifespan. Researchers used neural networks to analyze data and identify the impact of a previously overlooked organ. The findings could influence future approaches to healthcare and preventative medicine.
Five Papers Offer Clear Insights Into Large Language Models
A recent roundup highlights five research papers that effectively explain large language models (LLMs) to a broad audience. The papers cover core concepts underpinning LLMs and help demystify their operations, making advanced AI topics more accessible.