KLT-based quality controlled compression of a single lead egg

• US 6,152,883 A
• Filed: 06/23/1999
• Issued: 11/28/2000
• Est. Priority Date: 06/23/1998
• Status: Expired due to Fees

• Current Assignee: Dalhousie University
• Original Assignee: Dalhousie University
• Inventors: Blanchett, Travis Paul, Fenton, Gordon Ashley, Kember, Guy Cecil
• Primary Examiner(s): O'Connor, Cary
• Assistant Examiner(s): Natnithithadha, Navin
• Application Number:
  US09/338,828
• Time in Patent Office: 524 Days
• Field of Search: 600/508, 600/509, 600/515, 600/516, 600/517, 600/521
• US Class Current: 600/521
• CPC Class Codes: A61B 5/349 Detecting specific parameters of the electrocardiograph cycle
• CPC Class Codes: A61B 5/7232 involving compression of the physiological signal, e.g. to extend the signal recording period
• CPC Class Codes: A61B 5/7253 characterised by using transforms

Abstract

A compression algorithm for the compression of ECG recordings uses the Karhunen-Loeve Transform (KLT) to transform a set of N sampled ECG beats from a matrix of N×M samples into a new form, from which a selected subset can be retained for storage, transmission, or analysis. In order to reduce computation time and storage space, a multirate downsampling operation may be applied, which retains the appropriate spectral information in each block. The downsampled beats are then padded to make them of uniform size, and a Karhunen-Loeve Transform is applied to the sample set. Coefficients from the Karhunen-Loeve Transform of the sample set are retained for reconstruction according to one of two criteria. The average variance of the reconstructed sample set may be controlled, or different numbers of coefficients may be retained for each beat. The KLT compressed data may be reconstituted by reverse KLT transforming the data.

Claims

1. A method of encoding a plurality of sampled ECG records, each record representing a heartbeat, said method comprising the steps of:

(a) aligning the records about their R-peaks;

(b) sizing the records to have the same number of samples and forming a sample set of the sized records; and

(c) performing a Karhunen-Loeve Transform (KLT) on the sample set, thereby generating a set of eigenvalues and associated eigenvectors for the sample set, and a set of KLT coefficients for each record.

2. The method of claim 1, further including the steps of:

(d) ordering the eigenvalues and their associated eigenvectors in decreasing order of eigenvalue size; and

(e) retaining an ordered subset of the most significant KLT coefficients for each record and an ordered subset of the eigenvectors associated with largest eigenvalues for the sample set.

3. The method of claim 2, wherein the retaining step (e) further retains the m largest eigenvalues for the sample set.

4. The method of claim 3, wherein the retaining step (e) retains the same number m of KLT coefficients for each record and m eigenvectors and eigenvalues for the sample set.

5. The method of claim 3, wherein the retaining step (e) retains a variable number of KLT coefficients for each record, according to a criterion which minimizes reconstruction error.

6. The method of claim 2, wherein the retaining step (e) retains the same number m of KLT coefficients for each record and m eigenvectors for the sample set.

7. The method of claim 2, wherein the retaining step (e) retains a variable number of KLT coefficients for each record, according to a criterion which minimizes reconstruction error.

8. The method of claim 7 further including the step of flagging the records for which the most coefficients are retained, for analysis.

9. The method of claim 2, wherein the step of sizing (b) the records comprises the step of padding the ends of the shorter records.

10. The method of reconstituting a desired set of ECG records encoded according to the method of claim 2, comprising the steps of:

retrieving the retained ordered KLT coefficients for each record of the desired set and the retained ordered eigenvectors for the desired set; and

performing a reverse Karhunen-Loeve Transform (KLT) on the retrieved KLT coefficients and eigenvectors to generate a plurality of records.

11. The method of claim 2 further including the step of retaining the actual number of samples in each record prior to sizing each record.

12. The method of reconstituting a desired set of ECG records encoded according to the method of claim 11, comprising the steps of:

retrieving the retained ordered KLT coefficients for each record, actual number of samples for each record, and ordered eigenvectors for the desired set;

performing a reverse Karhunen-Loeve Transform (KLT) on the retrieved KLT coefficients and eigenvectors to generate a plurality of records; and

resizing each record according to the retrieved actual number of samples for that record.

13. The method of claim 2, further including the step of downsampling each record prior to forming the sample set.

14. The method of claim 13 wherein each record is downsampled at multiple rates, and wherein the downsampling is more severe at slowly varying portions of the record and less severe at quickly varying portions of the record.

15. The method of claim 14 wherein the step of downsampling at multiple rates downsamples the PQ and ST portions of the record at least about twice as heavily as the QRS portion of the record.

16. The method of reconstituting a desired set of ECG records encoded according to the method of claim 15, comprising the steps of:

retrieving the retained ordered KLT coefficients for each record, the actual number of samples for each record, and the ordered eigenvectors for the desired set;

performing a reverse Karhunen-Loeve Transform (KLT) on the retrieved KLT coefficients and eigenvectors to generate a plurality of records; and

upsampling each record at multiple rates selected to restore the original sample rate of each portion of the record.

17. Apparatus for encoding a plurality of sampled ECG records, each record representing a heartbeat, said apparatus comprising:

means for centering the records about their R-peaks;

means for sizing the records to have the same number of samples and forming a sample set of the sized records; and

means for performing a Karhunen-Loeve Transform (KLT) on the sample set, thereby generating a set of eigenvalues and eigenvectors for the sample set, and a set of KLT coefficients for each record.

18. The apparatus of claim 17, further comprising:

means for ordering the eigenvalues and their associated eigenvectors in decreasing order of eigenvalue size; and

means for retaining an ordered subset of the most significant KLT coefficients for each record and an ordered subset of the eigenvectors associated with largest eigenvalues for the sample set.

19. The apparatus of claim 18, wherein the means for retaining retains the same number m of KLT coefficients for each record and retains the m eigenvectors associated with the m largest eigenvalues for the sample set.

20. The apparatus of claim 18, wherein the means for retaining retains a variable number of KLT coefficients for each record, according to a criterion which minimizes reconstruction error, and retains enough eigenvectors associated with the largest eigenvalues to reconstitute the record for which the largest number of KLT coefficients are retained.

21. The apparatus of claim 20, further including means for flagging the records for which the most coefficients are retained, for analysis.

22. The apparatus of claim 18, further including means for downsampling each record at multiple rates prior to forming the sample set.

23. Apparatus for reconstituting a desired set of ECG records encoded by the apparatus of claim 18, comprising:

means for retrieving the retained ordered KLT coefficients for each record in the desired set and the retained ordered eigenvectors for the desired set; and

means for performing a reverse Karhunen-Loeve Transform (KLT) on the retrieved KLT coefficients and eigenvectors to generate a plurality of records.

24. A method of detecting irregular heartbeats from a string of sampled heartbeat records comprising the steps of:

(a) centering the records about their R-peaks;

(b) sizing the records to have the same number of samples and forming a sample set of the sized records;

(c) performing a Karhunen-Loeve Transform (KLT) on the sample set, thereby generating a set of eigenvalues and associated eigenvectors for the sample set, and a set of KLT coefficients for each record;

(d) ordering the eigenvalues and their associated eigenvectors in decreasing order of eigenvalue size;

(e) retaining an ordered subset of the most significant KLT coefficients for each record and an ordered subset of the eigenvectors associated with largest eigenvalues for the sample set wherein a variable number of KLT coefficients for each record is retained, according to a criterion which minimizes reconstruction error; and

(f) flagging each record for which more than a predetermined number of coefficients are retained, as pertaining to potentially irregular heartbeats.

25. The method of claim 24, further including the step of:

sounding an alarm if any records are flagged by step (f).

26. The method of claim 24, further including the step of:

applying medication if any records are flagged by step (f).

27. The method of claim 24, further including the step of:

reconstituting any flagged records by applying a reverse KLT to the coefficients for that record and the eigenvectors for the sample set; and

displaying the reconstituted beats for analysis.

28. The method of claim 24, wherein the sampled heartbeat records are retreived from storage.

29. The method of claim 24, wherein the sampled heartbeat records are sampled in real time from a patient'"'"'s monitored heartbeats.