A Machine-Learning Framework to Identify Distinct Phenotypes of Aortic Stenosis Severity

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Aortic stenosis (AS) symptoms arise from valvular obstruction and heart muscle strain.
• Current echocardiography (ECHO) for AS severity is limited by diagnostic uncertainty.

Purpose of the Study:

• To develop and validate machine learning (ML) models for enhanced echocardiographic grading of AS severity.
• To improve the classification of AS severity and optimize aortic valve replacement (AVR) timing.

Main Methods:

• Echocardiography (ECHO) data from 1,052 patients were used for patient similarity analysis to define AS phenogroups.
• A supervised ML classifier was developed and validated using computed tomography (CT) and cardiovascular magnetic resonance (CMR) imaging cohorts.
• Prognostic value was assessed using clinical outcomes like AVR and death.

Main Results:

• ML classified 57% of 1,964 patients as high-severity AS.
• High-severity AS patients showed increased valve calcification (CT) and left ventricular mass/fibrosis (CMR).
• ML classification improved discrimination and reclassification for AVR outcomes compared to conventional methods.

Conclusions:

• Machine learning effectively integrates ECHO data to augment AS severity classification.
• This ML approach has significant potential for optimizing AVR timing in AS patients.
• ML-based classification provides prognostic value, especially in subgroups with nonsevere or discordant AS.