부산대학교 도서관

Antiarrhythmic drug choice for atrial fibrillation (AF) neglects the individual ionic current profile of the patient, even though it determines drug safety and efficacy. We hypothesize that the electrocardiogram (ECG) might contain information critical for pharmacological treatment personalization. Thus, this study aims to identify the extent of atrial ionic information embedded in the ECG, using multi-scale modeling and simulation. A dataset of 1,000 simulated ECGs was computed using a population of human-based whole-atria models with 200 individual ionic profiles and 5 different torso-atria orientations. A regression neural network was built to predict key atrial ionic conductances based on P- and $T_{a^{-}}$ wave biomarkers. The neural network predicted, with >80% precision, the density of seven ionic currents relevant for AF, namely, ultra-rapid $(I_{Kur})$, rapid $(I_{Kr})$, outward transient $(I_{to})$, inward rectifier $K^{+}(I_{Kl})$, L-type $Ca^{2+}(I_{CaL}),Na^{+}/K^{+}pump (I_{NaK})$ and fast $Na^{+}(I_{Na})$ currents. These ionic densities were identified through the $P-(i.e.,I_{Na}),T_{a^{-}}(i.e.,\ I_{Kl},\ I_{NaK})$ or both waves $(i.e.,\ I_{Kur},\ I_{Kr},\ I_{to},\ I_{CaL})$, providing a non-invasive characterization of the atrial electrophysiology. This could improve patient stratification and cardiac safety and the efficacy of AF pharmacological treatment.