OR WAIT null SECS
Model could improve outcomes, reduce costs and save lives, researchers say.
An artificial intelligence (AI) neural network identified congestive heart failure with 100% accuracy, according to the findings of a study published in Biomedical Signal Processing and Control Journal.
Just one raw electrocardiogram (ECG) heartbeat was what the AI needed to identify the condition, according to the paper.
“Enabling clinical practitioners to access an accurate (congestive heart failure) detection tool can make a significant societal impact, with patients benefiting from early and more efficient diagnosis and easing pressures on (National Health Service) resources,” said Leandro Pecchia, Ph.D., assistant professor of biomedical engineering at the University of Warwick in England.
Typical congestive heart failure detection methods focus on heart variability and are time consuming and prone to errors, according to researchers.
Instead, the research team developed a model which uses a combination of advanced signal processing and machine-learning tools on raw ECG signals.
Researchers performed an analysis on two public data sets. Data for the control group were from the MIT-BIH Normal Sinus Rhythm Database. This database included 18 long-term ECG recording of normal health not-arrhythmic subjects. The data from the congestive heart failure group were from the BIDMC Congestive Heart Failure Database. The data included long-term ECG recordings of 15 subjects with severe congestive heart failure, according to the study.
Researchers labeled each heartbeat with a one or zero according to whether the subject was healthy or suffering from congestive heart failure. The data were split into three subsets for training, validation and testing.
The machine-learning model was trained and evaluated more over 10 runs.
The model was evaluated based on individual beat classification and classification on five minutes of ECG excerpts through a majority voting scheme.
Researchers computed accuracy, precision, sensitivity, specificity and area under the curve to assess the model.
The model had a specificity and sensitivity of approximately 96%, the researchers reported.
“Our model delivered 100% accuracy — by checking just one heartbeat we were able to detect whether a person had heart failure,” said Sebastiano Massaro, Ph.D., associate professor of organizational neuroscience and behavioral science at the University of Surrey in England.
The research team encourages future research and for clinical practices to apply the framework. This will allow for further validation and could enable moving from retrospective detection of congestive heart failure to prospective diagnosis, the researchers suggest. This could improve well-being, reduce healthcare costs and potentially save lives, the research team concluded.
The study is titled “A convolutional neural network approach to detect congestive heart failure” and can be accessed here.
Get the best insights in digital health directly to your inbox.