An electrocardiogram-based AI algorithm for early detection of pulmonary hypertension

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Early diagnosis of pulmonary hypertension (PH) is crucial for effective management.
• Pulmonary hypertension poses significant health risks if not identified promptly.

Purpose of the Study:

• To develop and validate an AI algorithm for PH screening using standard 12-lead ECG.
• To assess the AI's ability to detect PH early, prior to clinical diagnosis.

Main Methods:

• A convolutional neural network, the PH Early Detection Algorithm (PH-EDA), was trained on retrospective ECG data.
• The algorithm classified patients as PH-likely or PH-unlikely based on ECG features.
• External validation was performed using independent datasets from Mayo Clinic and Vanderbilt University Medical Center.

Main Results:

• The PH-EDA achieved high performance, with an AUC of 0.92 at Mayo Clinic and 0.88 at VUMC on diagnostic test sets.
• The algorithm demonstrated effectiveness in pre-emptive screening, with AUCs of 0.86 and 0.81, respectively.
• Significant detection capability persisted up to 5 years prior to diagnosis, with AUCs remaining above 0.73.

Conclusions:

• The PH-EDA can effectively screen for pulmonary hypertension using ECG data.
• The algorithm shows potential for accelerating PH diagnosis and management.
• AI-powered ECG analysis offers a promising avenue for early detection of cardiovascular diseases.