Artificial Intelligence Diagnosis of Paroxysmal Atrial Fibrillation Using Transthoracic Echocardiography During Sinus Rhythm

Background: Atrial fibrillation (AF) often goes undetected due to its paroxysmal and asymptomatic nature. This study aimed to evaluate the capability of deep neural networks (DNN) in identifying occult AF using transthoracic echocardiography (TTE) conducted during sinus rhythm, and to assess the prognostic implications of AI-driven predictions on patient outcomes.

Methods: Our cohort comprised 85,686 TTEs from Sheba Medical Center, spanning the years 2009 to 2021. A DNN algorithm developed by Aisap.ai was utilized to diagnose paroxysmal AF using A4C and PLAX views without Doppler imaging. Ground truth for underlying AF was established through ICD codes, ECGs, or Holter monitors, with the earliest detection date considered the onset. The cohort was divided into a training set (60%), a validation set (20%), and a test set (20%). Model diagnostic performance was evaluated. The primary clinical endpoint comprised cerebrovascular events, heart failure hospitalizations, and all-cause mortality. Clinical outcomes were assessed with patients classified into four groups: false positives (FP), false negatives (FN), true positives (TP), and true negatives (TN).

Results: The model demonstrated robust diagnostic metrics, achieving an area under the curve of 0.85, sensitivity of 0.8, and specificity of 0.72. The test set included 17,140 patients with a median age of 66 (IQR 54-75), consisting of 9,857 (57.6%) males and 3,449 (20%) with paroxysmal AF. It estimated that 6,581 (38.4%) of patients had paroxysmal AF, of which 3,832 (58%) were previously undiagnosed. During a median follow-up of 6.5 (IQR 4-9.5) years, 4,981 (29%) patients experienced the composite endpoint. A multivariate analysis showed that patients classified as AF by the model without prior diagnosis (FP), were 24% more likely to meet the study outcome compared to those in sinus rhythm (TN) (95% CI 1.13-1.35, p<0.001).

Conclusion: Incorporating AI into standard TTE analysis improves the detection of underlying AF in sinus rhythm patients. This could enhance the identification of individuals at risk for adverse cardiovascular events, warranting further monitoring for AF diagnostics.