STATISTICAL ASSESSMENT AND CALIBRATION OF NUMERICAL ECG MODELS

Objective: Because of the inter-subject variability of ECGs in a healthy population, it is not straightforward to assess the quality of synthetic ECGs produced by deterministic mathematical models. We propose a statistical method to address this question. Methods: We use a dataset of 1588 healthy, real ECGs and we introduce a way to calibrate the deterministic model so that its output fits the dataset. Our approach is based on the concepts of spatial quantiles and spatial depths. These notions are convenient to manipulate functional data since they provide a nonparametric way to measure the discrepancy of the model output with a distribution of data. Results: The method is successfully applied to two very different models: a phenomenological model based on ordinary differential equations, and a complex biophysical model based on partial differential equations set on a threedimensional geometry of the heart and the torso. We show in particular that the proposed calibration strategy allows us to improve the quality of the ECG obtained with the biophysical model. Significance: The proposedmethodology is to our knowledge the first attempt to assess the quality of synthetic ECGs with quantitative statistical arguments. More generally it can be applied to other situations where a deterministic model produces a functional output that has to be compared with a population of measurements containing inter-subject variability.