Echocardiographic view classification using ECG-assisted multi-phase image sampling

Transthoracic echocardiography (TTE) is the most common non-invasive imaging technique used to diagnose cardiac disease, capturing video assessments from up to 27 unique views. Accurate view classification is crucial for downstream tasks using TTE data. Traditional methods often rely on random or fixed video frame sampling, leading to misaligned training data. This study introduces a novel deep learning approach that uses pixel-embedded ECG signals to enhance TTE view classification. We developed an ensemble model using three cardiac phase-specific images, sampled from: R-wave(early-systole), late-systole, and diastole, each trained as CNN models based on the EfficientNet-B2. A majority vote was used to ensemble the results of each individual model prediction and achieved high performance, with accuracy of 92.4% and AUC of 99%, across 11 TTE views, offering a robust and efficient solution for classification. Further validation is needed to confirm generalizability across larger cohorts and variable data inputs.