Recognition of individual heart rate patterns with cepstral vectors

Abstract.

 Heart rate patterns may contain diagnostic as well as forensic information. To test these possibilities, individual heart rate patterns were represented as heart-rate cepstral vectors (HRCVs) computed in 12 dimensions via linear predictive coding (LPC) of brief segments of heart rate. A library of codebook vectors was computed for 12 cardiac patients from a standard ECG database. Statistical classification of subjects was based on the minimal weighted distances between test and codebook vectors. Weights were based on the ratio of inter- to intrasubject variances of their cepstral coefficients. Results showed that: (1) HRCV coefficients adequately reproduced the HRV spectrum, and (2) HRCV distances could be used to identify individuals within the group with a reliability of 93%. Thus, heart rate variations are an individual characteristic that can be represented as a single 12-dimensional vector.