PULSE-HF: A Deep Learning Model to Predict Heart Failure
Heart failure, a chronic and incurable disease characterized by weakened or damaged heart muscles, leads to a gradual buildup of fluid in a patient’s lungs, legs, feet, and other body parts. Despite advancements in treatments involving healthy lifestyle changes, prescription medications, and pacemakers, heart failure remains a leading cause of morbidity and mortality worldwide. A study published in Lancet eClinical Medicine by a team of researchers from MIT, Mass General Brigham, and Harvard Medical School, presents a promising development in predicting heart failure. The team has developed a deep learning model, PULSE-HF, that accurately predicts changes in left ventricular ejection fraction (LVEF), a key indicator of heart health.
Understanding PULSE-HF
PULSE-HF, which stands for “predicting changes in left ventricular systolic function from ECGs of patients with heart failure,” was developed and tested retrospectively on three different patient cohorts. These include Massachusetts General Hospital, Brigham and Women’s Hospital, and MIMIC-IV, a publicly available dataset.
Healthy human hearts pump out about 50 to 70 percent of the blood from the left ventricle with each beat. Anything less is considered a sign of possible heart failure. According to Tiffany Yau, an MIT graduate student and co-first author of the PULSE-HF paper, the model only needs one electrocardiogram (ECG) to predict whether a patient’s ejection fraction will fall below 40 percent within the next year. This indicates the most severe subset of heart failure.
Benefits of PULSE-HF
If PULSE-HF predicts that a patient’s ejection fraction is likely to worsen within a year, the doctor can prioritize the patient for follow-up care. This allows lower-risk patients to reduce the number of hospital visits and the time it takes to attach 10 electrodes to their body for a 12-lead ECG. The model can also be used in low-resource clinical settings, such as medical practices in rural areas that do not typically employ a cardiac sonographer to perform daily ultrasound examinations.
What sets PULSE-HF apart is its predictive capabilities. Unlike other heart failure ECG methods, it is not intended for detection, but rather for prediction. There are currently no other methods to predict future LVEF decline in heart failure patients.
Performance of PULSE-HF
During the testing and validation process, researchers used a metric called area under the receiver operating characteristic curve (AUROC) to measure the performance of PULSE-HF. AUROC is typically used to measure a model’s ability to distinguish between classes on a scale of 0 to 1, where 0.5 is random and 1 is perfect. PULSE-HF achieved AUROCs ranging from 0.87 to 0.91 in all three patient cohorts.
Remarkably, researchers have also developed a version of PULSE-HF for single ECGs, meaning only one electrode needs to be attached to the body. While 12-lead ECGs are generally considered superior because they are more comprehensive and accurate, the performance of the single-channel version of PULSE-HF was just as strong as the 12-lead version.
Challenges and Prospects
Despite the promising results, developing PULSE-HF was not without its challenges. One of the team’s biggest challenges was collecting, processing, and cleaning the ECG and echocardiogram datasets. Despite these hurdles, researchers remain optimistic about the model’s potential. The next step for PULSE-HF will be to test the model in a prospective study on real patients whose future ejection fraction is unknown.
Despite the challenges of getting clinical AI tools like PULSE-HF across the finish line, the researchers believe that the years of hard work have been worth it. The development and implementation of AI in healthcare stand as a testament to the potential of technology in addressing pressing health issues like heart failure, ultimately working towards alleviating suffering and improving patient care.
Source: MIT News
