System and method for interpreting electrocardiograms

• US 7,996,073 B2
• Filed: 10/13/2004
• Issued: 08/09/2011
• Est. Priority Date: 10/13/2004
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Busche, Frederick D., Kelkar, Bhooshan P.
• Primary Examiner(s): Patel; Niketa I
• Assistant Examiner(s): Lavert; Nicole F
• Application Number:
  US10/963,968
• Publication Number:
  US 20060079795A1
• Time in Patent Office: 2,491 Days
• Field of Search: 600509-510, 600512-513, 600516-519, 600/507, 600/9, 607/14, 607/28
• US Class Current: 600/513
• CPC Class Codes: A61B 5/349 Detecting specific parameters of the electrocardiograph cycle

Abstract

A system and method for interpreting electrocardiogram data. A system is provided that clusters raw electrocardiogram (EKG) data into clusters of EKG data; generates a predictive model for each cluster of EKG data; compares inputted patient EKG data with the clusters of EKG data to identify a matching cluster of EKG data; applies the predictive model associated with the matching cluster of EKG data to the inputted patient EKG data; and outputs diagnostic data.

Claims

1. A computer-implemented electrocardiogram (EKG) diagnosis system, comprising:

a sequence analysis system for identifying EKG data patterns in raw EKG data and clustering the EKG data patterns into clusters of EKG data patterns,

wherein the EKG data patterns include a PR interval, a PR segment, an ST segment, a QT interval, and P, QRS, T, and U waves,

wherein each of the clusters comprises a plurality of EKG data patterns, and wherein each of the EKG data patterns within a cluster shares a similar pattern signature, and has a dissimilar pattern signature from EKG data patterns outside the cluster;

a predictive analysis system for generating a predictive model for each cluster of EKG data patterns;

a system for associating a diagnosis with each cluster, wherein the diagnosis is one of a healthy condition or one of at least one predefined health condition,

wherein the predictive model provides a probability determined based on a radial basis function that the pattern signature of each of the plurality of clusters is representative of the diagnosis; and

a system for outputting diagnostic data for each cluster of EKG data patterns, wherein the diagnostic data includes the probability that the pattern signature of each of the plurality of clusters is representative of the diagnosis.

2. The EKG diagnosis system of claim 1, wherein the sequence analysis system includes a pattern time series algorithm for evaluating similarities among clusters of EKG data patterns and identifying and clustering EKG patterns having similar data patterns.

3. The EKG diagnosis system of claim 1, wherein the sequence analysis system includes a comparing system for comparing inputted patient EKG data patterns with the clusters of EKG data patterns to identify a matching cluster of EKG data patterns.

4. The EKG diagnosis system of claim 3, wherein the predictive analysis system includes a system for applying the predictive model associated with the matching cluster of EKG data patterns to the inputted patient EKG data pattern, and for outputting diagnostic data.

5. The EKG diagnosis system of claim 4, wherein diagnostic data includes a diagnosis and a probability associated with the diagnosis.

6. The EKG diagnosis system of claim 2, wherein the pattern time series algorithm comprises one of:

an algorithm that identifies sequences having a same value or a similar value within a threshold at a corresponding time,

an algorithm that uses a scale invariant analysis to identify sequences that share a similar trend but are of a different scale,

an algorithm that uses a subsequence matching analysis to identify sequences that are similar despite differences in slope, and

an algorithm that uses a shift in time analysis to identify sequences that are time shifted relative to one another, but are otherwise similar.

7. A program product stored on a recordable non-transitory storage medium for interpreting electrocardiogram (EKG) data comprising:

means for identifying EKG patterns in raw EKG data patterns in raw EKG data and clustering the EKG data patterns into clusters of EKG patterns, wherein the EKG patterns include a PR interval, a PR segment, an ST segment, a QT interval, and P, QRS, T, and U waves, wherein each of the clusters comprises a plurality of EKG patterns, and wherein each of the EKG data patterns within a cluster shares a similar pattern signature, and has a dissimilar pattern signature from EKG data patterns outside the cluster;

means for generating a predictive model for each cluster of EKG data patters;

means for associating a diagnosis with each cluster, wherein the diagnosis with each cluster, wherein the diagnosis is one of a healthy condition or one of at least one predefined health condition, wherein the predictive model provides a probability determined based on a radial basis function that the pattern signature of each of the plurality of clusters is representative of the diagnosis; and

means for outputting diagnostic data for each cluster of EKG data patterns, wherein the diagnostic data includes the probability that the pattern signature of each of the plurality of clusters is representative of the diagnosis.

8. The program product of claim 7, wherein the clustering means includes a pattern time series algorithm for evaluating similarities among clusters of EKG data patterns and identifying and clustering EKG patterns having similar data patterns.

9. The program product of claim 7, further comprising means for comparing inputted patient EKG data patterns with the clusters of EKG data patterns to identify a matching cluster of EKG data patterns.

10. The program product of claim 9, further comprising means for applying the predictive model associated with the matching cluster of EKG data patterns to the inputted patient EKG data patterns, and for outputting diagnostic data.

11. The program product of claim 10, wherein diagnostic data includes a diagnosis and a probability associated with the diagnosis.

12. The program product of claim 8, wherein the pattern time series algorithm comprises one of:

an algorithm that identifies sequences having a same value or a similar value within a threshold at a corresponding time,

an algorithm that uses a scale invariant analysis to identify sequences that share a similar trend but are of a different scale,

an algorithm that uses a subsequence matching analysis to identify sequences that are similar despite differences in slope, and

an algorithm that uses a shift in time analysis to identify sequences that are time shifted relative to one another, but are otherwise similar.

13. A method for interpreting electrocardiogram (EKG) data, comprising:

identifying EKG data patterns in raw EKG data and clustering the EKG data patterns into clusters of EKG data patterns,

wherein the EKG data patterns include a PR interval, a PR segment, an ST segment, a QT interval, and P, QRS, T, and U waves,

wherein each of the clusters comprises a plurality of EKG data patterns, and wherein each of the EKG data patterns within a cluster shares a similar pattern signature, and has a dissimilar pattern signature from EKG data patterns outside the cluster;

generating a predictive model for each cluster of EKG data patterns; and

associating a diagnosis with each cluster, wherein the diagnosis is one of a healthy condition or one of at least one predefined health condition, and wherein the predictive model provides a probability determined based on a radial basis function that the pattern signature of each of the plurality of clusters is representative of the diagnosis, and

outputting diagnostic data for each cluster of EKG data patterns, wherein the diagnostic data includes the probability that the pattern signature of each of the plurality of clusters is representative of the diagnosis.

14. The method of claim 13, wherein the clustering step utilizes a pattern time series algorithm for evaluating similarities among clusters of EKG data patterns and identifying and clustering EKG patterns having similar data patterns.

15. The method of claim 13, further comprising the steps of:

comparing inputted patient EKG data patterns with the clusters of EKG data patterns to identify a matching cluster of EKG data patterns; and

applying the predictive model associated with the matching cluster of EKG data patterns to the inputted patient EKG data patterns.

16. The method of claim 14, wherein the pattern time series algorithm comprises one of:

an algorithm that identifies sequences having a same value or a similar value within a threshold at a corresponding time,

an algorithm that uses a scale invariant analysis to identify sequences that share a similar trend but are of a different scale,

an algorithm that uses a subsequence matching analysis to identify sequences that are similar despite differences in slope, and

an algorithm that uses a shift in time analysis to identify sequences that are time shifted relative to one another, but are otherwise similar.

17. A method for deploying an application for interpreting electrocardiogram (EKG) data, comprising:

providing a computer infrastructure being operable to;

identify EKG data patterns in raw EKG data and cluster the EKG data patterns into clusters of EKG data patterns,

wherein the EKG data patterns include a PR interval, a PR segment, an ST segment, a QT interval, and P, QRS, T, and U waves,

wherein each of the clusters comprises a plurality of EKG data patterns, and wherein each of the EKG data patterns within a cluster shares a similar pattern signature, and has a dissimilar pattern signature from EKG data patterns outside the cluster;

generate a predictive model for each cluster of EKG data patterns;

compare inputted patient EKG data patterns with the clusters of EKG data patterns to identify a matching cluster of EKG data patterns;

associate a diagnosis with each cluster, wherein the diagnosis is one of a healthy condition or one of at least one predefined health condition;

apply the predictive model associated with the matching cluster of EKG data patterns to the inputted patient EKG data patterns, wherein the predictive model provides a probability determined based on a radial basis function that the pattern signature of each of the plurality of clusters is representative of the diagnosis; and

output diagnostic data for each cluster of EKG data patterns, wherein the diagnostic data includes the probability that the pattern signature of each of the plurality of clusters is representative of the diagnosis.

18. A method of interpreting electrocardiogram data, the method comprising:

identifying EKG data patterns in raw EKG data and clustering the EKG data patterns into clusters of EKG data patterns,

wherein the EKG data patterns include a PR interval, a PR segment, an ST segment, a QT interval, and P, QRS, T, and U waves,

wherein each of the clusters comprises a plurality of EKG data patterns, and wherein each of the EKG data patterns within a cluster shares a similar pattern signature, and has a dissimilar pattern signature from EKG data patterns outside the cluster;

generating a predictive model for each cluster of EKG data patterns;

comparing inputted patient EKG data patterns with the clusters of EKG data patterns to identify a matching cluster of EKG data patterns;

associating a diagnosis with each cluster, wherein the diagnosis is one of a healthy condition or one of at least one predefined health condition;

applying the predictive model associated with the matching cluster of EKG data patterns to the inputted patient EKG data patterns, wherein the predictive model provides a probability determined based on a radial basis function that the pattern signature of each of the plurality of clusters is representative of the diagnosis; and

outputting diagnostic data for each cluster of EKG data patterns, wherein the diagnostic data includes the probability that the pattern signature of each of the plurality of clusters is representative of the diagnosis.