PMC

Normalized histogram of the NIAP (black) and fitted gamma distribution (dark blue) calculated from 250 times reshuffling of the beats within an analysis window (L = 64 beats). Marked are the 95^th and 99^th percentiles of the calculated alternans amplitude (non-parametric) and the fitted gamma distribution (parametric). If the estimated TWA is larger than the 95^th or 99^th percentile one can confidently reject the null hypothesis at α = 0.05 or α = 0.01, respectively.

Range of NIAP amplitudes at 95^th percentile significance in reshuffled beat sequences using the SAM (see ), representing a statistical measure of the upper-limit on the NIAP. Median (boxes), 5% (lower line) and 95% (upper line) are plotted to illustrate spread of the 99th percentile at each simulated TWA amplitude and across all simulated records, at SNR of 10 dB and no noise scenario.

Performance of the MMA method (top) and the SM (bottom) using a k threshold value of 3. Note that, such choice of k results in rejection of almost 98 ± 2% of false estimates at 0 μV TWA amplitudes and rejection of 50 ± 20% of estimated values at the largest simulated TWA amplitude and SNR of 10. See for explanation of legend.

Examples of simulated VCG with TWA amplitude of 23 μV. Physiological noise consisting of a mixture of muscle artifacts, electrode movements, and baseline wander are added to each record. Only simulations at SNR of 10 dB (top) and clean VCG (bottom) are shown here. Zooming into the bottom plot one can observe the micro-volt variations from a normal beat (type-A) to an abnormal beat (type-B). The maximum amplitude variation between a type-A and a type-B beat is concentrated around the T-wave peak.

Comparison of the NSRDB, CHFDB, and SCDDB patient populations (at matched HRs) using the SAM (top) and the SDA method (bottom). The small numbers by the open blue circles indicate the number of detected episodes of TWA for the given HR range. The grey error bars signify the percentage of indeterminate cases at each HR range over the entire population. Note that, in the top panels the indeterminate cases are caused by preprocessing failure of associated analysis windows, while in the indeterminate cases in the bottom panels are an aggregate result of preprocessing failure and application of the SDA method (α = 0.05). In comparison to the top panels, the number of detected episodes of TWA in the bottom panels are greatly reduced (see numbers by the open blue circles), and the margin of separability among patient populations is increased (see ).

Performance of the SAM on the simulated data before (top) and after (bottom) application of the SDA method (α = 0.01). Estimated TWA amplitude (Calc. TWA Amp.) versus simulated TWA amplitude at SNR of 10 dB (left) and noise free simulated VCGs (right) are shown. Each point on the figure is calculated from 50 simulated VCG records of 5 minutes length each. At a heart rate of 70 ± 5 beats/min this results in roughly 10 TWA amplitude measurements per record, and thus a total of 500 estimates. Represented in each figure are the lower 5%, median, and the upper 95% of the estimated TWA amplitude, as well as, the line y = x (representing ideal detection). The grey color error bars represent the percentage of indeterminate cases (%indet.). Note that, at the SNR of 10 dB application of surrogate testing resulted in rejection of approximately 99% of episodes around 0μV simulated TWA amplitude, and simultaneously the percentage of missed detections (or indeterminate cases) at higher TWA amplitudes is notably smaller than the SM method (see ).