Detection of atrial septal aneurysm on ECG based on Deep Learning algorithm (ANN)

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Atrial Septal Aneurysm (ASA) presents diagnostic challenges due to nonspecific symptoms.
• ASA diagnosis is frequently incidental, lacking specific electrocardiogram (ECG) criteria.
• Understanding ASA's clinical significance is limited.

Purpose of the Study:

• To evaluate the efficacy of Machine Learning (ML) in detecting Atrial Septal Aneurysm (ASA) from ECG data.
• To develop and validate an ML model for ASA identification.
• To explore ML's potential in improving ASA diagnosis.

Main Methods:

• Retrospective analysis of 233 individuals (123 with ASA, 110 without).
• Trans-thoracic Echocardiography (TTE) confirmed ASA presence.
• An Artificial Neural Network (ANN) was trained and tested on ECG parameters.

Main Results:

• The ANN model achieved 73% sensitivity and 84% specificity.
• Positive Predictive Value (PPV) was 80%, Negative Predictive Value (NPV) was 73%, and F-1 score was 0.79.
• Area Under the Curve (AUC) of 0.8 indicated excellent diagnostic performance.

Conclusions:

• ML, specifically ANN, demonstrates feasibility for detecting ASA via ECG.
• This approach offers a potential non-invasive method for identifying ASA.
• Further research can enhance clinical understanding and management of ASA.