AI automated heart failure detection and classification from electronic health records

New research led by the University of Dundee, demonstrated the feasibility of AI to automatically identify and classify patients with heart failure  from archived echocardiographic images with  Us2.ai software.

Heart failure (HF) is a highly prevalent yet under-diagnosed condition with high mortality and morbidity. Echocardiography is a foundational investigation to diagnose HF and differentiate the types of HF i.e. HF with reduced (HFrEF), mildly reduced (HFmrEF) and preserved (HFpEF) ejection fraction.

Electronic health records (EHRs) are an increasingly high-quality data source that can be used for the creation of pragmatic cohort studies, disease surveillance, case selection for clinical trials (RCTs), and quality improvement initiatives. The quality and quantity of EHR data are expanding and increasingly include EHR-linked biobanks and EHR-linked imaging data.

This study aimed to identify and classify patients with HF from routinely stored EHR data, linked to Scottish Health Research Register bioresource and echocardiographic data collected from the Tayside and Fife region of Scotland using a deep learning-based approach.