Augmented detection of septal defects using advanced optical coherence tomography network-processed phonocardiogram

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Cardiac auscultation is a primary method for identifying congenital heart disease (CHD), but can fail with faint murmurs or obesity.
• Accurate diagnosis of CHD, including atrial septal defects (ASDs) and ventricular septal defects (VSDs), is crucial for patient outcomes.

Purpose of the Study:

• To develop and evaluate an intelligent diagnostic method for detecting heart murmurs associated with ASDs and VSDs.
• To compare the diagnostic performance of the developed AI model against expert cardiologist assessment.

Main Methods:

• Digital heart sound and phonocardiogram recordings were collected from 184 participants.
• An advanced optical coherence tomography network (AOCT-NET) was utilized to classify phonocardiogram data (normal, ASD, VSD) after feature extraction.
• Performance was evaluated by comparing AOCT-NET results with echocardiography and cardiologist assessments.

Main Results:

• The AOCT-NET demonstrated superior sensitivity (76.4%) and specificity (90%) in detecting ASDs compared to cardiologists (27.8% sensitivity, 98.5% specificity).
• The AOCT-NET showed no significant advantage over cardiologists in detecting VSDs.
• Echocardiography confirmed 88 healthy participants, 50 with ASDs, and 46 with VSDs.

Conclusions:

• The AOCT-NET shows potential for improving ASD detection rates, particularly in screening asymptomatic individuals.
• This AI-driven approach may serve as a valuable tool for early detection of certain congenital heart conditions.