Identification of Fiducial Points in Serial Data

Abstract

A method is described for fiducial point identification that can be tuned to specific data types using training set data having manually marked fiducial points. The role of the “expert” in the training process is limited to providing the correct point identification in the training data. No articulation of the mathematical justification for the choice is needed. The method is based on the calculation of a weighted score for each point in an unknown data record. The score is derived from a doubly normalized computation of values for a set of generic discriminant functions. Candidate points are identified by an order and selection process. Because of the multiple normalization and use of sorting for selection, the method is independent of scale or range of the data to be identified. Neither the training process nor the identification process requires any dimensional input, other than the identification of fiducial points for use in the training process. Examples are given using cardiac electrogram data and ultrasonic data.