WIT Press

Statistical Parameter Estimation And Signal Classification In Cardiovascular Diagnosis


Free (open access)

Paper DOI






Page Range

457 - 469




3139 kb


S. Bernhard, K. Al Zoukra & C. Schütte


Medical technology has seen impressive success in the past decades, generating novel clinical data at an unexpected rate. Even though numerous physiological models have been developed, their clinical application is limited. The major reason for this lies in the difficulty of finding and interpreting the model parameters, because most problems are ill-posed and do not have unique solutions. On the one hand the reason for this lies in the information deficit of the data, which is the result of finite measurement precision and contamination by artifacts and noise and on the other hand on data mining procedures that cannot sufficiently treat the statistical nature of the data. Within this work we introduce a population based parameter estimation method that is able to reveal structural parameters that can be used for patient-specific modeling. In contrast to traditional approaches this method produces a distribution of physiologically interpretable models defined by patient-specific parameters and model states. On the basis of these models we identify disease specific classes that correspond to clinical diagnoses, which enable a probabilistic assessment of human health condition on the basis of a broad patient population. In an ongoing work this technique is used to identify arterial stenosis and aneurisms from anomalous patterns in parameter space. We think that the information-based approach will provide a useful link between mathematical models and clinical diagnoses and that it will become a constituent in medicine in near future. Keywords: statistical cardiovascular system model, cardiovascular system identification, multi-channel measurement, state-space model, parameter estimation, Bayesian signal classification, patient-specific diagnosis.


statistical cardiovascular system model, cardiovascular systemidentification, multi-channel measurement, state-space model, parameterestimation, Bayesian signal classification, patient-specific diagnosis