Computer versus cardiologist: Is a machine learning algorithm able to outperform an expert in diagnosing a...

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Phospholamban (PLN) p.Arg14del mutations are linked to dilated/arrhythmogenic cardiomyopathy.
• Electrocardiogram (ECG) features are crucial for diagnosing PLN p.Arg14del cardiomyopathy.
• Machine learning (ML) shows promise in ECG analysis, potentially exceeding human expert performance.

Purpose of the Study:

• Develop and evaluate ML and deep learning models for diagnosing PLN p.Arg14del cardiomyopathy using ECGs.
• Compare the diagnostic accuracy of these models against expert cardiologists.

Main Methods:

• Trained ML models on ECG data from 155 PLN mutation carriers and 155 controls.
• Utilized 4-fold cross-validation for model development and testing.
• Validated the best models on an external dataset (n=50) and compared performance with expert cardiologists.

Main Results:

• ML models achieved higher sensitivity (0.65-0.81) and accuracy (range not specified, but superior to experts) than cardiologists (0.64 and 0.28, respectively).
• Expert cardiologists showed higher specificity (0.99) compared to ML models (0.53-0.81).
• T-wave morphology was identified as the most critical feature for classifying PLN p.Arg14del carriers.

Conclusions:

• ML models demonstrate superior diagnostic performance over expert cardiologists for PLN p.Arg14del cardiomyopathy.
• ECG T-wave morphology is a significant diagnostic indicator for this condition.