부산대학교 도서관

Background. Larger one-time values of spatial QRS-T angle (SA) are associated with risk. However, experience how serial changes in SA $(\Delta SA)$ should be interpreted is lacking. Even within normal limits, any $\Delta SA$ likely signifies electrical remodeling. This study aimed to assess the impact of choosing either $\Delta SA$ or $\vert \Delta SA\vert$ as one of a set of serial ECG difference features that constitute the input for our deep learning serial-ECG classifier (DLSEC). Methods. DLSEC was trained and tested to detect emerging pathology in two serial ECG databases: a heart failure database and an acute ischemia database. Either $\Delta SA$ or $\vert \Delta SA\vert$ were among 13 features of serial-ECG differences. DLSEC was dynamically generated during learning, and testing area under the curve (AUC) of the receiver operating characteristic was computed. Results. The DLSECs performed well in emerging heart failure as well as in acute ischemia: testing AUCs were 72% and 84% for the heart failure database and 77% and 83% for the ischemia database, for $\Delta SA$ or $\vert \Delta SA\vert$ among the features, respectively. Conclusion. $\vert \Delta SA\vert$ among the features was superior to $\Delta SA$ in discriminating cases and controls. Our study supports the concept that any $\Delta SA$, irrespective of its sign, indicates a worsening clinical condition. Further corroboration requires studies in other clinical situations.