Introduction: Personalisation of cardiac electrophysiology models remains an important challenge for guiding atrial fibrillation treatment. Our aim was to calibrate tissue parameters using synthetic measurements of local activation times (LAT) and/or action potential durations (APD) on the left atrium (LA), assuming homogeneous parameter fields, and is a first step towards calibration of realistic heterogeneous models using clinically accessible data.
Methods: We used a monodomain model with excitability described by the modified Mitchell-Schaeffer (mMS) model. S1S2 simulations were performed using 202 latin hypercube samples of the model parameters to obtain LAT and APD with 3xS1 beats at 800ms cycle length followed by an S2 beat with a coupling interval of 500ms. These data were used to train Gaussian process emulators for LAT and APD at a few LA locations. The emulators were then used for calibration against synthetic measurements via a novel adaptation of the ensemble Kalman filter. Calibration used 50 ground truths, each for three types of synthetic measurements: LAT for the third beat (S1), LAT for the last two beats (S1+S2), and APD in addition to two LATs (S1+S2+APD).
Results: We were able to calibrate two mMS parameters, Tau_in and conductivity, from LAT data alone. A third parameter, Tau_open, was difficult to calibrate from S1 LAT data alone but we found improved calibration using S1+S2 LAT, and further using S1+S2 LAT and APD (Figure 1, left). The effect of improved Tau_open calibration on error in predicted APD is shown in Figure 1, right.
Conclusion: We have shown that calibration of a simple and phenomenological model from LAT measured at a small number of locations with an S1S2 protocol is feasible using our novel method. Measurements of APD (or a surrogate such as effective refractory period) would further improve the calibration.