Effects of Prior Data on the Inference and Filtering Based Electrocardiographic Imaging

Statistical estimation techniques with "good" prior information improves the accuracy of electrocardiographic imaging (ECGI). Obtaining a "good" prior information in terms of training data, and how to use these data to estimate the prior parameters, are among of the primary chall...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:Computing in cardiology s. Page 1 - Page 4
Hlavní autoři: Erenler, Taha, Dogrusoz, Yesim Serinagaoglu
Médium: Konferenční příspěvek
Jazyk:angličtina
Vydáno: Creative Commons 01.09.2019
Témata:
Kalman filters
Maximum likelihood estimation
Noise measurement
Parameter estimation
Signal to noise ratio
Training
ISSN:2325-887X
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:Statistical estimation techniques with "good" prior information improves the accuracy of electrocardiographic imaging (ECGI). Obtaining a "good" prior information in terms of training data, and how to use these data to estimate the prior parameters, are among of the primary challenges in statistical ECGI literature. This study investigates the effects of training set compositions, and prior parameter estimation methods on the ECGI accuracy. Two different training sets were used to determine the prior statistical parameters: 1) Beats that are all paced close to the test beat pacing site, 2) Beats with pacing sites covering a wider region around the test beat pacing site. These two training sets are obtained from a database of previously recorded epicardial potentials and used in maximum likelihood (ML)- and maximum a posteriori (MAP)-based prior estimation methods. The inverse problem is then solved by using the Kalman filter, based on those priors. Our results show that the Tikhonov regularization is the most fragile method to the measurement noise. MAPIF method is more robust to measurement noise in terms of electrogram reconstruction and activation time estimation accuracies, for both training sets. MLIF performed better with a more spread training set in terms of electrogram reconstruction accuracy, but showed better activation time reconstruction performance with the first training set.
ISSN:2325-887X
DOI:10.22489/CinC.2019.292