Feature-based expert system classifier

• US 5,819,007 A
• Filed: 03/15/1996
• Issued: 10/06/1998
• Est. Priority Date: 03/15/1996
• Status: Expired due to Term

• Current Assignee: Draeger Medical Systems Incorporated (FPS Vermögensverwaltung GmbH)
• Original Assignee: Siemens Medical Systems, Inc. (Siemens AG)
• Inventors: Elghazzawi, Ziad F.
• Primary Examiner(s): Downs, Robert W.
• Application Number:
  US08/616,362
• Time in Patent Office: 935 Days
• Field of Search: 395/61, 395/51, 382/228, 128/702, 128/705, 607/14
• US Class Current: 706/46
• CPC Class Codes: A61B 5/349 Detecting specific parameters of the electrocardiograph cycle
• CPC Class Codes: A61B 5/363 Detecting tachycardia or bradycardia
• CPC Class Codes: A61B 5/364 Detecting abnormal ECG interval, e.g. extrasystoles, ectopic heartbeats
• CPC Class Codes: A61B 5/366 Detecting abnormal QRS complex, e.g. widening
• CPC Class Codes: A61B 5/7264 Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems neural networks per se G06N3/00; expert systems per se G06N5/00
• CPC Class Codes: A61B 5/7267 involving training the classification device
• CPC Class Codes: G16H 50/20 for computer-aided diagnosis, e.g. based on medical expert systems
• CPC Class Codes: G16H 50/70 for mining of medical data, e.g. analysing previous cases of other patients
• CPC Class Codes: Y10S 706/924 Medical

Abstract

A method and apparatus for classifying an input as belonging to one of a plurality of predefined classes, comprises: (a) developing a plurality of feature values and forming a feature vector from the feature values which is representative of the input, (b) applying the feature vector to a knowledge-base comprising predictivity curves for each class of said plurality of classes, and developing a plurality of predictivity values for each feature, each predictivity value being indicative of a likelihood of the input belonging to a respective one of each of said classes based on the feature value; (c) combining the predictivity values developed for each of the features for each class to generate a total predictivity value for each class; and (d) generating a determination of class based upon the total predictivity values generated by the prior step.

Claims

1. A method for classifying an input as belonging to one of a plurality of predefined classes, comprising:

(a) developing a plurality of feature values and forming a feature vector from the feature values which is representative of the input;

(b) directly deriving curves of predictivity values from a reference database of inputs having pre-existing class annotations, to produce a knowledge base of principally learned classification information;

(c) applying the feature vector to the knowledge-base for developing a plurality of predictivity values for each feature, each predictivity value being indicative of a likelihood of the input belonging to a respective one of each of said classes based on the feature value;

(d) combining the predictivity values developed for each of the features for each class to generate a total predictivity value for each class; and

(e) generating a determination of class based upon the total predictivity values generated by the prior step.

2. The method according to claim 1, wherein said determination of class is made by comparing the total positive predictivity value determined for each class and selecting as the determined class that class having the greatest positive predictivity value.

3. The method according to claim 1, further including the steps of:

developing a normal value for at least one of the features using prior values of said feature that have previously been classified as normal; and

developing a normalized feature for use in step (c) above, using the normal value determined for that feature.

4. The method according to claim 1, wherein the input is representative of a waveform.

5. The method according to claim 4, wherein one of said features is a shape feature which is compared with a predefined waveform shape for each of the classes and the result of the comparison is used as a feature in step (c).

6. The method according to claim 4, wherein said input comprises an EKG waveform of a patient being monitored, and said feature vector comprises a plurality of values representative of measurements of various attributes of said EKG waveform.

7. The method according to claim 6, wherein said various attributes of said EKG waveform comprise at least beat width, beat area and RR interval.

8. The method according to claim 6, wherein said various attributes of said EKG waveform comprise normalized measurements of the attributes, said normalization being a comparison of the attributes values of the current EKG waveform with the current dominant normal values for these attributes.

9. The method according to claim 6, wherein said input also includes as feature values measurement values of various attributes of a heartbeat signal generated using pulse oximetry.

10. The method according to claim 6, wherein said input also includes as feature values measurement values of various attributes of a heartbeat signal generated using a blood pressure measurement technique.

11. The method according to claim 6, wherein in step 1(e) said class determination is an EKG rhythm, such as normal sinus rhythm, bigeminy, trigeminy, fibrillation and tachycardia.

12. The method according to claim 4, wherein step 1(e) determines a class from the group of classes including at least normal and ventricular.

13. The method according to claim 4, wherein in step 1(e) said class determination is an alarm for a patient monitoring apparatus, and said input vector is representative of a plurality of physiological conditions of a patient being monitored, said conditions including EKG rhythm, blood pressure measurements and pulse oximetry measurements.

14. The method according to claim 4 for use in a patient monitoring system, wherein in step 1(d) said class determination is used as a validation of the measurement of a physiological condition of a patient being monitored.

15. The method according to claim 14, wherein said validation comprises the classification of the input as normal, and is used to validate the determination of the pulse rate and oxygen saturation of said patient using pulse oximetry.

16. The method according to claim 14, wherein said validation comprises the classification of the input into a known classification, such as as normal or abnormal, and is used to validate the determination of blood pressure values obtained from a patient using an invasive blood pressure monitoring technique.

17. The method according to claim 1, wherein in step (c) the predictivity value is modified based upon the certainty of the feature value being accurate.

18. The method according to claim 1, wherein in step (c) the predictivity value used for a given feature value is determined not only by the value of the feature, but also on the existence of an external condition.

19. The method according to claim 18, wherein said feature value is a measurement of the RR interval in a given portion of an EKG waveform, and said external condition is the presence of an atrial fibrillation rhythm in that portion of the EKG waveform.

20. The method according to claim 1, wherein step 1(d) develops a normalized summation of the predictivity values developed for each feature of each class.

21. The method according to claim 20, wherein step 1(d) develops said normalized summation in accordance with the following equation:

space="preserve" listing-type="equation">cfh=cfh+pph(k)*(1-cfh), for k=1;

N,

where cfh=0 upon initialization, and

where N represents the total amount of evidences(features).

22. An apparatus for classifying an input as belonging to one of a plurality of predefined classes, comprising:

extracting means responsive to an input signal for extracting a plurality of features therefrom and developing therefore a corresponding plurality of feature values, and forming a feature vector from the feature values which is representative of the input signal;

a learning means for directly deriving curves of predictivity values from a reference database of inputs having pre-existing class annotations, thereby producing a knowledge-base of principally learned classification information;

applying means for applying the feature vector to the knowledge base for developing at its output a plurality of predictivity values for each feature value of the feature vector, each predictivity value being indicative of a likelihood of the input belonging to a respective one of each of said classes based on the feature value;

a combining means responsive to said plurality of predictivity values for combining the predictivity values developed for each of the features for each class, and generating a total predictivity value for each class; and

determining means for generating a determination of class based upon the total predictivity values generated by the combining means.

23. Apparatus according to claim 22, wherein said determining means includes comparing means for comparing the total predictivity value determined for each class and selecting as the determined class that class having the greatest predictivity value.

24. Apparatus according to claim 22, wherein said input signal comprises an EKG waveform of a patient being monitored, and said extracting means develops as said feature vector a plurality of values representative of measurements of various attributes of said EKG waveform.

25. Apparatus according to claim 24, wherein said extracting means includes means for developing values for at least beat width, beat area and RR interval.

26. Apparatus according to claim 24, further including normalization means for developing normalized measurements of the attributes of said EKG waveform, said normalization means including means for comparing the attribute values of the current EKG waveform with average normal values for these attributes.

27. Apparatus according to claim 22, wherein said combining means comprises a summation means for developing a normalized summation of the predictivity values developed for each feature of each class.

28. Apparatus according to claim 27, wherein said combining means develops said normalized summation in accordance with the following equation:

space="preserve" listing-type="equation">cfh=cfh+pph(k)*(1-cfh), for k=1;

N,

where cfh=0 upon initialization, and

where N represents the total amount of evidences(features).