Non-invasive prediction of risk for sudden cardiac death
A method and apparatus for the quantitative determination of an individual's risk for sudden cardiac death (SCD) is described. Risk stratification is accomplished (and may have a sensitivity and specificity of greater than about 90%) by determining the presence in any individual being tested for SCD risk of sequences identified herein to correlate quantitatively with SCD risk. Both the number of such sequences present and their alignment scores (similarity) with the SCD risk sequence ensemble are used to calculate quantitative SCD risk.
1. A method for determining, isolating, and constructing an optimized collection of digital electrocardiogram (ECG) sequences for quantitative stratification of sudden cardiac death (SCD) risk, the method comprising:
receiving ECG measurements taken from a first group of individuals who have no history of heart disease and no history of SCD;
receiving ECG measurements taken from a second group of individuals who have a history of any type of heart disease but no history of SCD;
receiving ECG measurements taken from a third group of individuals who have a history of SCD;
identifying, by a processor, at least one ECG sequence from ECG measurements relatively unique to members of the third group of individuals;
wherein the identifying comprises isolating, and optimally grouping with a processor, ECG sequences indicative of SCD risk, the isolating comprising, in a processor, assessing quality of alignment of ECG sequences, first derivatives of the ECG sequences and second derivatives of the ECG sequences from multiple individuals of the groups; and
creating, in a memory, a collection of the identified ECG sequences.
2. The method of claim 1 , wherein a preprocessor preprocesses each digital ECG measurement by:
denoising the digital ECG measurement; and
calculating first and second derivatives from the denoised ECG data.
3. The method of claim 2 , wherein the digital ECG measurement is denoised by at least one of a Finite Impulse Response (FIR) filter and a wavelet denoising method employing entropy calculations to optimize threshold settings.
4. The method of claim 3 , wherein the noise includes at least one of electrical noise, mechanical noise, respiration artifacts, white noise, movement artifact, and baseline drift.
5. The method of claim 1 , wherein the processor aligns each ECG sequence present from each individual of the third group of individuals with ECG sequences present from each individual of the first, second and third groups of individuals,
determines an alignment score for each ECG sequence relatively unique to members of the third group of individuals compared with members of the first and second groups of individuals, and
determines whether an ECG sequence correlates with a risk for SCD based on its presence and alignment error score in any individual of the first, second and third groups of individuals.
6. The method of claim 1 , wherein the processor is configured to:
align the digital ECG sequences present in each individual of the third group of individuals with the digital ECG sequences present in each individual of the first, second, and third groups of individuals to obtain alignments,
determine an alignment error score for each of the alignments,
using the alignment error scores, identify ECG sequences relatively unique to members of the third group of individuals,
determine which of the ECG sequences is correlated with a risk for SCD based upon their presence and results of clinical data.
7. The method of claim 1 , further comprising:
testing accuracy of the ECG sequences indicative of the risk for SCD.
8. The method of claim 1 , wherein each digital ECG measurement is obtained from a standard resting digital 12-lead ECG.