Cramér-Rao Lower Bounds of Model-Based Electrocardiogram Parameter Estimation

Clinical parameter estimation from the electrocardiogram (ECG) is a recurrent field of research. It debated that ECG parameters estimated by human experts and machines/algorithms always model-based (implicitly or explicitly). Therefore, all algorithm used in this context have performance bounds terms achievable mean squared error, which are not exceedable. These depend on adopted data-model, scheme (least-squares maximum likelihood, Bayesian), prior assumptions model noise distributions. In research, we develop comprehensive theoretical framework for derive Cramér-Rao lower (CRLBs) most popular signal models modeling literature, namely functional expansions (including polynomials) sum Gaussian functions. The developed evaluated over real synthetic data three applications: T-to-R wave ratio estimation, ST-segment analysis QT-interval using state-of-the-art estimators each context. proposed derived CRLBs provide practical guidelines selection data-models, sampling frequency (beyond Nyquist rate), segment length, number beats required beat averaging, other factors influence accuracy ECG-based clinical estimation.