Neural network IDs congestive heart failure with 100% accuracy

Researchers have successfully trained a convolutional neural network (CNN) to detect congestive heart failure (CHF) with 100% accuracy using data from just one heartbeat.

Sebastiano Massaro, an associate professor at the University of Surrey, and colleagues at the University of Warwick and University of Florence leveraged electrocardiogram (ECG) data to identify individual cases of CHF, applying AI in the process. They outlined their findings in Biomedical Signal Processing and Control.

According to the paper, the team’s goal was to develop a CNN model that could accurately detect CHF on the basis on a singular raw ECG heartbeat, juxtaposing existing methods grounded in heart rate variability. Right now, CHF detection is a pricey process associated with high mortality rates and sustained healthcare costs.

“Thus, with a worldwide aging population and sustained pressures on healthcare systems and resources, there is the compelling demand—among patients, healthcare providers, policymakers and the society as a whole—to address this scenario by identifying highly accurate methods to improve detection of heart failures and in turn enable early and more efficient diagnoses,” Massaro and co-authors wrote.

The team trained and tested their CNN model on publicly available ECG datasets comprising a total of 490,505 heartbeats—some healthy and some with CHF. Current methods of CHF diagnosis are time-consuming and prone to error; Massaro et al.’s approach delivered 100% accuracy in a much tighter time frame.

“By checking just one heartbeat we are able to detect whether or not a person has heart failure,” he said in a release. “Our model is also one of the first known to be able to identify the ECG’s morphological features specifically associated to the severity of the situation.”

In the journal, Massaro and co-authors said they’re hopeful that their novel model will be used as a generalizable classification method for other classification tasks in medicine and other fields. Before that can happen, though, more work needs to be done to expand and confirm the team’s results.