We applied multiresolution wavelet analysis to the sequence of times betweenhuman heartbeats (R-R intervals) and have found a scale window, between 16 and32 heartbeats, over which the widths of the R-R wavelet coefficients fall intodisjoint sets for normal and heart-failure patients. This has enabled us tocorrectly classify every patient in a standard data set as either belonging tothe heart-failure or normal group with 100% accuracy, thereby providing aclinically significant measure of the presence of heart-failure from the R-Rintervals alone. Comparison is made with previous approaches, which have provided onlystatistically significant measures.
展开▼