Method and apparatus for detecting a condition associated with acute cardiac ischemia

• US 6,171,256 B1
• Filed: 12/11/1998
• Issued: 01/09/2001
• Est. Priority Date: 04/30/1998
• Status: Expired due to Fees

• Current Assignee: Physio-Control Incorporated (Stryker Corporation)
• Original Assignee: Physio-Control Corporation
• Inventors: Schmitt, Paul W., Joo, Tae H., Hampton, David R., Medema, Douglas K., Niskanen, Robert A.
• Primary Examiner(s): EVANISKO, GEORGE ROBERT
• Application Number:
  US09/209,879
• Time in Patent Office: 760 Days
• Field of Search: 600/508, 600/509, 600/512, 600/515, 600/516
• US Class Current: 600/508
• CPC Class Codes: A61B 5/30 Input circuits therefor
• CPC Class Codes: A61B 5/308 for electrocardiography [ECG]
• CPC Class Codes: A61B 5/327 Generation of artificial ECG signals based on measured signals, e.g. to compensate for missing leads
• CPC Class Codes: A61B 5/35 by template matching
• CPC Class Codes: A61B 5/7267 involving training the classification device
• CPC Class Codes: G16H 50/70 for mining of medical data, e.g. analysing previous cases of other patients

Abstract

A device (10) that detects and reports the presence of an acute cardiac ischemic condition in a patient includes sensing electrodes (12, 14, 16, 18, 20, 22, 24, 26, 28, and 30) placed on a patient for acquiring multiple (e.g., twelve) lead ECG data from the patient. The device evaluates the acquired ECG data by analyzing global features derived from the ECG data. Global features include projection coefficients calculated from projecting a concatenated vector of representative heartbeat data onto sets of basis vectors that define an acute cardiac ischemic ECG subspace and a non-ischemic ECG subspace. A classifier evaluates the global features to determine whether an acute cardiac ischemic condition is detected. In a further embodiment of the invention, one or more classifiers evaluate local features, such as local morphological features and patient clinical information, in addition to the global features to determine whether an acute cardiac ischemic condition is detected. The device (10) may be configured with an adjustable sensitivity/specificity operating point. The result of the evaluation is reported to the user of the device.

Claims

1. A method of detecting and reporting a condition associated with acute cardiac ischemia in a patient, comprising:

(a) obtaining one or more leads of ECG data from the patient;

(b) deriving heartbeat data from the patient'"'"'s ECG data;

(c) forming a vector of heartbeat data from the derived heartbeat data;

(d) producing a set of global features by projecting the vector of heartbeat data onto one or more basis vectors that define an acute cardiac ischemic ECG subspace or a non-ischemic ECG subspace;

(e) classifying the set of global features to determine whether the global features are indicative of an acute cardiac ischemic condition; and

(f) reporting whether the acute cardiac ischemic condition is determined to be present.

2. The method of claim 1, wherein forming the vector of heartbeat data includes:

(a) analyzing the one or more leads of ECG data to identify one or more heartbeats;

(b) generating representative heartbeat data for each lead; and

(c) concatenating the representative heartbeat data for each lead to form the vector of heartbeat data.

3. The method of claim 1, wherein producing the set of global features includes:

(a) calculating an inner product of the vector of heartbeat data and one or more basis vectors that define the acute cardiac ischemic ECG subspace to produce a corresponding number of ischemic condition projection coefficients;

(b) calculating an inner product of the vector of heartbeat data and one or more basis vectors that define the non-ischemic ECG subspace to produce a corresponding number of non-ischemic condition projection coefficients; and

(c) using the ischemic condition projection coefficients and the non-ischemic condition projection coefficients as the set of global features.

4. The method of claim 1, further comprising:

(a) defining a plurality of groups wherein each group has basis vectors associated therewith that define a group-specific acute cardiac ischemic ECG subspace and a group-specific non-ischemic ECG subspace;

(b) categorizing the patient'"'"'s ECG data into a group in the plurality of groups; and

(c) using the basis vectors of the group into which the patient'"'"'s ECG data is categorized as the basis vectors onto which the vector of heartbeat data is projected.

5. The method of claim 4, wherein categorizing the patient'"'"'s ECG data into a group includes:

(a) selecting a local feature derived from the patient; and

(b) categorizing the patient'"'"'s ECG data into a group based on the selected local feature.

6. The method of claim 4, further comprising:

(a) defining each group of the plurality of groups to correspond to a location of an acute cardiac ischemic condition; and

(b) if the acute cardiac ischemic condition is determined to be present, reporting the location of the acute cardiac ischemic condition corresponding to the group into which the patient'"'"'s ECG data is categorized.

7. The method of claim 4, wherein categorizing the patient'"'"'s ECG data into a group includes:

(a) calculating the ST elevation of one or more of the leads obtained from the patient;

(b) forming subgroups of the leads for which ST elevation was calculated;

(c) calculating a composite ST elevation for each subgroup by calculating a mathematical combination of the ST elevation of the leads in each subgroup; and

(d) categorizing the patient'"'"'s ECG into a group according to the subgroup whose composite ST elevation is greatest.

8. The method of claim 4, further comprising:

(a) measuring an ST elevation on the one or more leads of ECG data obtained from the patient; and

(b) categorizing the patient'"'"'s ECG data into a group based on the measured ST elevation.

9. The method of claim 1, further comprising:

(a) using a Karhunen-Loeve transformation to calculate a first set of basis vectors that define the acute cardiac ischemic ECG subspace and a second set of basis vectors that define the non-ischemic ECG subspace;

(b) selecting one or more basis vectors from the first set of basis vectors and one or more basis vectors from the second set of basis vectors as the basis vectors onto which the vector of heartbeat data is projected.

10. The method of claim 1, wherein classifying the set of global features includes:

(a) concatenating the set of global features to form a global feature vector;

(b) producing a classification statistic by evaluating the global feature vector relative to a representative global feature vector derived from a training population; and

(c) comparing the classification statistic with a threshold to determine whether the global features are indicative of the acute cardiac ischemic condition.

11. The method of claim 6, wherein a Gaussian classifier is used to evaluate the global feature vector relative to the representative global feature vector and produce the classification statistic.

12. The method of claim 10, further comprising selecting the threshold in accordance with a desired sensitivity/specificity tradeoff.

13. The method of claim 1, wherein if the one or more basic vectors define only the acute cardiac ischemic ECG subspace, or if the one or more basis vectors define only the non-ischemic ECG subspace, then classifying the set of global features includes:

(a) concatenating the set of global features to form a global feature vector;

(b) producing a classification statistic based on a calculated distance between the global feature vector and a representative global feature vector derived from a training population; and

(c) comparing the classification statistic with a threshold to determine whether the global features are indicative of the acute cardiac ischemic condition.

14. A device for detecting and reporting a condition associated with acute cardiac ischemia in a patient, comprising:

(a) a plurality of electrodes for sensing patient ECG signals;

(b) a processing unit in communication with the plurality of electrodes the processing unit configured to;

(i) derive heartbeat data from the patient'"'"'s ECG data;

(ii) form a vector of heartbeat data from the derived heartbeat data;

(iii) produce a set of global features by projecting the vector of heartbeat data onto one or more predetermined basis vectors that define an acute cardiac ischemic ECG subspace or a non-ischemic ECG subspace; and

(iv) classify the set of global features to determine whether the global features are indicative of an acute cardiac ischemic condition; and

(c) a user output in communication with the processing unit for reporting whether the acute cardiac ischemic condition is determined to be present.

15. The device of claim 14, wherein the processing unit is configured to classify the set of global features to produce a first output, and wherein the processing unit is further configured to:

(a) obtain a set of local features from the patient;

(b) classify the set of local features obtained from the patient to produce a second output; and

(c) classify the first and second outputs to determine whether the first and second outputs are indicative of the acute cardiac ischemic condition.

16. The device of claim 14, wherein the processing unit is configured to obtain a set of local features from the patient and classify the set of local features in combination with the set of global features to determine whether the acute cardiac ischemic condition is present.

17. A method of detecting and reporting a condition associated with acute cardiac ischemia in a patient, comprising:

(a) obtaining one or more leads of ECG data from the patient;

(b) deriving heartbeat data from the patient'"'"'s ECG data;

(c) forming a vector of heartbeat data from the derived heartbeat data;

(d) generating a local classification statistic by;

(i) deriving a set of local features from the patient; and

(ii) classifying the set of local features to produce the local classification statistic;

(e) generating a global classification statistic by;

(i) producing a set of global features by projecting the vector of heartbeat data onto one or more basis vectors that define an acute cardiac ischemic ECG subspace or a non-ischemic ECG subspace; and

(ii) classifying the set of global features to produce the global classification statistic;

(f) classifying the local classification statistic and the global classification statistic to determine whether the local and global classification statistics are indicative of an acute cardiac ischemic condition; and

(g) reporting whether the acute cardiac ischemic condition is determined to be present.

18. The method of claim 17, wherein forming the vector of heartbeat data includes:

(a) analyzing the one or more leads of ECG data to identify one or more heartbeats;

(b) generating representative heartbeat data for each lead; and

(c) concatenating the representative heartbeat data for each lead to form the vector of heartbeat data.

19. The method of claim 17, wherein classifying the set of local features includes:

(a) concatenating the set of local features to form a local feature vector; and

(b) evaluating the local feature vector relative to a representative local feature vector derived from a training population to produce the local classification statistic.

20. The method of claim 17, wherein classifying the set of local features includes:

(a) applying one or more local features in the set of local features to a logistic regression equation to produce a probability of detection that is used as a composite local feature;

(b) dichotomizing the composite local feature; and

(c) classifying the dichotomized composite local feature in producing the local classification statistic.

21. The method of claim 17, wherein classifying the set of local features includes:

(a) concatenating the set of local features to form a local feature vector;

(b) calculating a Mahalanobis distance between the local feature vector and a representative local feature vector derived from a training population, wherein the Mahalanobis distance is used as a composite local feature; and

(c) classifying the composite local feature in producing the local classification statistic.

22. The method of claim 21, further comprising:

(a) dividing the training population into a plurality of groups;

(b) deriving a representative local feature vector for each group;

(c) calculating a Mahalanobis distance between the local feature vector and the representative local feature vector for each group in the plurality of groups, resulting in a plurality of composite local features; and

(d) classifying the plurality of composite local features in producing the local classification statistic.

23. The method of claim 17, wherein classifying the set of local features includes:

(a) defining a plurality of groups, wherein each group has a representative local feature vector associated therewith that is derived from a training population;

(b) concatenating the set of local features to form a local feature vector;

(c) for each group, calculating a Mahalanobis distance between the local feature vector and the representative local feature vector associated with the group;

(d) identifying the local feature vector with one of the groups based on the Mahalanobis distance calculated for each group, the group identification being used as a composite local feature; and

(e) classifying the composite local feature in producing the local classification statistic.

24. The method of claim 17, wherein classifying the set of local features includes:

(a) defining a plurality of groups wherein each group has a logistic regression equation associated therewith, the logistic regression equation for each group having regression coefficients derived from a logistic regression model using selected local features of a training population;

(b) applying corresponding selected local features derived from the patient to the logistic regression equation associated with each group to produce for each group a probability of detection that is used as a composite local feature; and

(c) classifying the composite local features in producing the local classification statistic.

25. The method of claim 17, wherein producing the set of global features includes:

(a) calculating an inner product of the vector of heartbeat data and one or more basis vectors that define the acute cardiac ischemic ECG subspace to produce a corresponding number of ischemic condition projection coefficients;

(b) calculating an inner product of the vector of heartbeat data and one or more basis vectors that define the non-ischemic ECG subspace to produce a corresponding number of non-ischemic condition projection coefficients; and

(c) using the ischemic condition projection coefficients and the non-ischemic condition projection coefficients as the set of global features.

26. The method of claim 17, further comprising:

(a) defining a plurality of groups wherein each group has basis vectors associated therewith that define a group-specific acute cardiac ischemic ECG subspace and a group-specific non-ischemic ECG subspace;

(b) categorizing the patient'"'"'s ECG data into a group in the plurality of groups; and

(c) using the basis vectors of the group into which the patient'"'"'s ECG data is categorized as the basis vectors onto which the vector of heartbeat data is projected.

27. The method of claim 26, wherein categorizing the patient'"'"'s ECG data into a group includes:

(a) selecting a local feature derived from the patient; and

(b) categorizing the patient'"'"'s ECG data into a group based on the selected local feature.

28. The method of claim 26, further comprising:

(a) measuring an ST elevation on the one or more leads of ECG data obtained from the patient; and

(b) categorizing the patient'"'"'s ECG data into a group based on the measured ST elevation.

29. The method of claim 26, further comprising:

(a) defining each group of the plurality of groups to correspond to a location of the acute cardiac ischemic condition; and

(b) if the acute cardiac ischemic condition is determined to be present, reporting the location of the acute cardiac ischemic condition corresponding to the group into which the patient'"'"'s ECG data is categorized.

30. The method of claim 26, wherein categorizing the patient'"'"'s ECG data into a group includes:

(a) calculating the ST elevation of one or more of the leads obtained from the patient;

(b) forming subgroups of the leads for which ST elevation was calculated;

(c) calculating a composite ST elevation for each subgroup by calculating a mathematical combination of the ST elevation of the leads in each subgroup; and

(d) categorizing the patient'"'"'s ECG into a group according to the subgroup whose composite ST elevation is greatest.

31. The method of claim 17, further comprising:

(a) using a Karhunen-Loeve transformation to calculate a first set of basis vectors that define the acute cardiac ischemic ECG subspace and a second set of basis vectors that define the non-ischemic ECG subspace;

(b) selecting one or more basis vectors from the first set of basis vectors and one or more basis vectors from the second set of basis vectors as the basis vectors onto which the vector of heartbeat data is projected.

32. The method of claim 17, wherein classifying the set of global features includes:

(a) concatenating the set of global features to form a global feature vector; and

(b) evaluating the global feature vector relative to a representative global feature vector derived from a training population to produce the global classification statistic.

33. The method of claim 32, wherein a Gaussian classifier is used to evaluate the global feature vector relative to the representative global feature vector and produce the global classification statistic.

34. The method of claim 17, wherein classifying the local and global classification statistics includes:

(a) producing a combined classification statistic by evaluating the local and global classification statistics relative to corresponding representative local and global classification statistics derived from a training population; and

(b) comparing the combined classification statistic with a threshold to determine whether the local and global classification statistics are indicative of the acute cardiac ischemic condition.

35. The method of claim 34, further comprising selecting the threshold in accordance with a desired sensitivity/specificity tradeoff.

36. A method of detecting and reporting a condition associated with acute cardiac ischemia in a patient, comprising:

(a) obtaining one or more leads of ECG data from the patient;

(b) deriving heartbeat data from the patient'"'"'s ECG data;

(c) forming a vector of heartbeat data from the derived heartbeat data;

(d) deriving a set of local features from the patient;

(e) producing a set of global features by projecting the vector of heartbeat data onto one or more basis vectors that define an acute cardiac ischemic ECG subspace or a non-ischemic ECG subspace;

(f) jointly classifying the set of global features and the set of local features to determine whether the global features and local features are indicative of an acute cardiac ischemic condition; and

(g) reporting whether the acute cardiac ischemic condition is determined to be present.

37. The method of claim 36, wherein forming the vector of heartbeat data includes:

(a) analyzing the one or more leads of ECG data to identify one or more heartbeats;

(b) generating representative heartbeat data for each lead; and

(c) concatenating the representative heartbeat data for each lead to form the vector of heartbeat data.

38. The method of claim 36, wherein the set of local features jointly classified with the set of global features includes a dichotomized composite local feature calculated by:

(a) applying one or more local features derived from the patient to a logistic regression equation to produce a probability of detection that is used as a composite local feature; and

(b) dichotomizing the composite local feature.

39. The method of claim 36, wherein the set of local features jointly classified with the set of global features includes a composite local feature calculated by concatenating the set of local features to form a local feature vector and calculating a Mahalanobis distance between the local feature vector and a representative local feature vector derived from a training population, the Mahalanobis distance being used as the composite local feature.

40. The method of claim 39, further comprising:

(a) dividing the training population into a plurality of groups;

(b) deriving a representative local feature vector for each group;

(c) calculating a Mahalanobis distance between the local feature vector and the representative local feature vector for each group, resulting in a plurality of composite local features; and

(d) including the plurality of composite local features in the set of local features jointly classified with the set of global features.

41. The method of claim 36, further comprising:

(a) concatenating the set of local features to form a local feature vector;

(b) defining a plurality of groups, and for each group, deriving a representative local feature vector from a training population;

(c) calculating a Mahalanobis distance for each group measured between the local feature vector and the representative local feature vector associated with each group;

(d) identifying the local feature vector with one of the groups based on the Mahalanobis distance calculated for each group, the group identification being used as a composite local feature; and

(e) jointly classifying the composite local feature with the set of global features and the set of local features to determine the presence of the acute cardiac ischemic condition in the patient.

42. The method of claim 36, wherein the set of local features jointly classified with the set of global features includes composite local features calculated by:

(a) defining a plurality of groups wherein each group has a logistic regression equation associated therewith, the logistic regression equation for each group having regression coefficients derived from a logistic regression model using selected local features of a training population; and

(b) applying corresponding selected local features derived from the patient to the logistic regression equation associated with each group to produce for each group a probability of detection that is used as a composite local feature.

43. The method of claim 36, wherein producing the set of global features includes:

(a) calculating an inner product of the vector of heartbeat data and one or more basis vectors that define the acute cardiac ischemic ECG subspace to produce a corresponding number of ischemic condition projection coefficients;

(b) calculating an inner product of the vector of heartbeat data and one or more basis vectors that define the non-ischemic ECG subspace to produce a corresponding number of non-ischemic condition projection coefficients; and

(c) using the ischemic condition projection coefficients and the non-ischemic condition projection coefficients as the set of global features.

44. The method of claim 36, further comprising:

(a) defining a plurality of groups wherein each group has basis vectors associated therewith that define a group-specific acute cardiac ischemic ECG subspace and a group-specific non-ischemic ECG subspace;

(b) categorizing the patient'"'"'s ECG data into a group in the plurality of groups; and

(c) using the basis vectors of the group into which the patient'"'"'s ECG data is categorized as the basis vectors onto which the vector of heartbeat data is projected.

45. The method of claim 44, wherein categorizing the patient'"'"'s ECG data into a group includes:

(a) selecting a local feature derived from the patient; and

(b) categorizing the patient'"'"'s ECG data into a group based on the selected local feature.

46. The method of claim 44, further comprising:

(a) defining each group of the plurality of groups to correspond to a location of the acute cardiac ischemic condition; and

(b) if the acute cardiac ischemic condition is determined to be present, reporting the location of the acute cardiac ischemic condition corresponding to the group into which the patient'"'"'s ECG data is categorized.

47. The method of claim 44, wherein categorizing the patient'"'"'s ECG data into a group includes:

(a) calculating the ST elevation of one or more of the leads obtained from the patient;

(b) forming subgroups of the leads for which ST elevation was calculated;

(c) calculating a composite ST elevation for each subgroup by calculating a mathematical combination of the ST elevation of the leads in each subgroup; and

(d) categorizing the patient'"'"'s ECG into a group according to the subgroup whose composite ST elevation is greatest.

48. The method of claim 44, further comprising:

(a) measuring an ST elevation on the one or more leads of ECG data obtained from the patient; and

(b) categorizing the patient'"'"'s ECG data into a group based on the measured ST elevation.

49. The method of claim 36, further comprising:

(a) using a Karhunen-Loeve transformation to calculate a first set of basis vectors that define the acute cardiac ischemic ECG subspace and a second set of basis vectors that define the non-ischemic ECG subspace;

(b) selecting one or more basis vectors from the first set of basis vectors and one or more basis vectors from the second set of basis vectors as the basis vectors onto which the vector of heartbeat data is projected.

50. The method of claim 36, wherein jointly classifying the set of global features and the set of local features includes:

(a) concatenating the set of local features and the set of global features to form a global/local feature vector; and

(b) producing a global/local classification statistic by evaluating the global/local feature vector relative to a representative global/local feature vector derived from a training population; and

(c) comparing the global/local classification statistic with a threshold to determine whether the global features and local features are indicative of the acute cardiac ischemic condition.

51. The method of claim 50, wherein a Gaussian classifier is used to evaluate the global/local feature vector relative to the representative global/local feature vector and produce the global/local classification statistic.

52. The method of claim 50, further comprising selecting the threshold in accordance with a desired sensitivity/specificity tradeoff.

53. A device for detecting and reporting a condition associated with acute cardiac ischemia in a patient, comprising:

(a) electrodes adapted to be placed on the patient to sense ECG signals;

(b) a user input;

(b) a processing unit in communication with the user input and the electrodes, wherein the processing unit is configured to acquire ECG data from the ECG signals and determine the presence of an acute cardiac ischemic condition by;

(i) analyzing the ECG data;

(ii) calculating a classification statistic reflective of a cardiac condition based on at least one characteristic obtained from the ECG data; and

(iii) comparing the classification statistic with a threshold that reflects a desired sensitivity/specificity operating point for the device, wherein the sensitivity/specificity operating point of the device is adapted to be selected by a user of the device via the user input;

wherein the threshold used by the processing unit is automatically adjusted for higher specificity after the processing unit determines the presence of the acute cardiac ischemic condition in the patient.

54. A device for detecting and reporting a condition associated with acute cardiac ischemia in a patient, comprising:

(a) electrodes adapted to be placed on the patient to sense ECG signals;

(b) a processing unit in communication with the electrodes, wherein the processing unit is configured to acquire ECG data from the ECG signals and determine the presence of an acute cardiac ischemic condition by;

(i) forming a vector of heartbeat data from the ECG data;

(ii) projecting the vector of heartbeat data onto one or more predetermined basis vectors defining an acute cardiac ischemic ECG subspace or a non-ischemic ECG subspace to produce one or more global features;

(iii) calculating a classification statistic reflective of a cardiac condition based on the one or more global features; and

(iv) comparing the classification statistic with a threshold that reflects a desired sensitivity/specificity operating point for the device.

55. The device of claim 54, wherein the device further includes a user input in communication with the processing unit and wherein the sensitivity/specificity operating point of the device is adapted to be selected by a user of the device via the user input.