Echocardiology benefits from an AI makeover

Echocardiograms use ultrasonic waves to reveal cardiac structure, function, and bloodflow through the heart and heart valves. Because they are portable and non-invasive, echocardiograms are widely used. However, in practice, data acquisition, processing, and interpretation are subjective and often inconsistent between practitioners.

Artificial intelligence techniques can aid in streamlining echocardiology to ensure more uniform application. Chang et al. investigated how machine learning and neural network approaches benefit echocardiology, discussing their potential opportunities and current challenges.

Currently, echocardiograms are administered by a highly trained operator, which adds additional cost and may be inaccessible for those in rural communities without fully equipped echocardiology laboratories. Instead, a machine learning-guided ultrasound probe could be used by practitioners without previous training.

Once the data is collected, deep learning algorithms can quickly quantify useful parameters like the heart chambers’ size, structure, and function.

Well-trained and validated convolutional neural networks (CNNs) can even aid in developing a diagnosis from the echocardiogram.

“Different CNNs have been developed for echocardiography image interpretation to distinguish visually similar conditions such as pathologic vs. physiologic hypertrophic cardiomyopathy, constrictive pericarditis vs. restrictive cardiomyopathy, and stress induced cardiomyopathy vs. acute myocardial infarction,” said author Kan Liu.

Though the authors caution that more training and clinical trials are needed before widespread implementation, they are hopeful that enabling easier use of echocardiograms means more people have access to them.

“Integrating novel deep learning models into regional imaging diagnostic networks may prompt a process of ‘democratization’ in access to echocardiograms, potentially reformatting a traditional echocardiography ‘lab’ operation to a contemporary echocardiography ‘system’ operation, reducing healthcare disparities in low resource or rural populations,” said Liu.

Source: “Deep learning from latent spatiotemporal information of the heart: Identifying advanced bioimaging markers from echocardiograms,” by Amanda Chang, Xiadong Wu, and Kan Liu, Biophysics Reviews (2024). The article can be accessed at https://doi.org/10.1063/5.0176850 .