B-PO02-181 THE COMPARISONS OF ARTIFICIAL INTELLIGENCE AND CARDIOLOGISTS FOR THE DIAGNOSIS OF TYPE 1 BRUGADA ELECTROCARDIOGRAM PATTERN

Abstract

An artificial intelligence (AI) system had been developed for the automatic diagnosis of the type 1 Brugada electrocardiogram (ECG) pattern in our previous report. To compare the accuracy of the diagnosis for the type 1 Brugada ECG pattern between the AI model and cardiologists. The 138 type 1 Brugada and 138 non-Brugada 12 lead ECGs were randomly selected from the ECG database of Taipei Veterans General Hospital. The ECG diagnoses of type 1 Brugada ECG patterns had been confirmed by two electrophysiologists as standard diagnoses based on 2017 AHA/ACC/HRS Guideline. All ECGs were blindly assigned to eight cardiologists and the AI model, respectively. The accuracy, sensitivity and specificity for the diagnosis of type I Brugada ECG pattern between AI and cardiologists were compared. The AUC curve of the ECG diagnosis of type 1 Brugada pattern was 0.96 (sensitivity: 88.4%, specificity: 89.1%). The average sensitivity of cardiologists was 62.7±17.8% while specificity was 98.5±3.0%. The prediction probability between the ECG diagnosis by the AI model and the standard diagnosis was highly consistent (Table, Kappa coefficient 0.78 and McNemar P=0.86). The predictive probability of cardiologists was only moderately consistent with the standard diagnosis. The accuracy of type 1 Brugada ECG patterns by the AI model was superior to those by the cardiologists. The cardiologists sometimes ignored the diagnosis of type 1 Brugada ECG, reflected by low sensitivity. The AI model was superior to the cardiologists for the diagnosis of type 1 Brugada ECG pattern. This implied the potential application of the AI model for early screening of Brugada syndrome.