Early prevention and detection of weak and malfunctioning hearts can undoubtedly help to reduce medical care costs. But how should the risks of patients with varying heart rate characteristics be used to estimate the likelihood of congestive heart failures? Shahbazi and Asl offer trusty algorithms for investigating the long-term variation in the features of heartbeat rates of patients with congestive heart failure (CHF), and for discriminating among patients subject to low and high risk of CHF. The reader unfamiliar with the use of the fast Fourier transform (FFT) method for estimating the power spectral density (PSD) should browse through its history [1], and the principles of discriminant algorithms [2], prior to exploring the unique algorithmic ideas in this paper.

The variation in the heartbeats of patients with heart conditions over time is the heart rate beat variance (HRBV). The authors used (1) the FFT method to estimate the linear PSD time and frequency domains of the HRBV features, (2) a generalized discriminant algorithm to extract the nonlinear features of the heart rate beat variance features (HRBVFs), and (3) a k-nearest neighbor algorithm to classify patients with CHF into low- and high-risk groups.

A reliable dataset mined from databases of patients with CHF was used to validate the accuracy of the algorithms proposed for estimating the risk of developing CHF. The fraction of all consecutive heartbeats over a time period was computed from the annotated heartbeats of patients with CHF in the experimental dataset. The authors then “mapped the input training data into a feature space in which samples are nonlinearly related to the input space, and then [applied] [linear discriminant analysis, LDA] ... to the mapped data.”

The authors accurately used the time domain and frequency domain of HRBVFs to explore the linear components of CHF. The forecast risks of developing CHF by use of the linear, nonlinear, and combined linear and nonlinear HRBVF algorithms are unbelievable compared to similar results in the literature. All researchers in bioinformatics and biomedicine should read and weigh in on the accuracy of the algorithms designed for assessing the risk of CHF in this paper.