Development and validation of AI-Enhanced auscultation for valvular heart disease screening through a multi-centre...

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Valvular heart disease (VHD) affects a growing population, with over half of cases undiagnosed due to late symptoms and limited screening.
• Current AI tools for VHD screening rely on murmur detection, which lacks sensitivity for certain VHD subtypes and is limited by small datasets.

Purpose of the Study:

• To develop and validate a novel neural network for direct prediction of clinically significant VHD from stethoscope recordings, using echocardiographic data for training.
• To improve VHD screening sensitivity and enable earlier diagnosis compared to existing methods.

Main Methods:

• A diverse dataset of 1767 patients from UK primary care and hospital settings was compiled, including stethoscope recordings and echocardiographic labels.
• A recurrent neural network was trained using echocardiographic targets, not murmur labels, to directly predict VHD.
• The algorithm's performance was evaluated using the area under the receiver operating characteristic curve (AUROC).

Main Results:

• The trained neural network achieved an AUROC of 0.83, outperforming general practitioners in VHD prediction.
• The algorithm demonstrated high sensitivity for severe aortic stenosis (98%) and severe mitral regurgitation (94%).
• The study utilized diverse patient data and validated the AI tool's effectiveness in real-world clinical settings.

Conclusions:

• The novel AI algorithm shows significant promise as a scalable, low-cost screening tool for valvular heart disease.
• Early detection of VHD through this AI-powered stethoscope analysis can facilitate timely referral and intervention.
• This approach represents a significant advancement in non-invasive VHD screening, addressing limitations of current methods.