Multi-domain motion estimation and plethysmographic recognition using fuzzy neural-nets

• US 6,931,269 B2
• Filed: 08/27/2004
• Issued: 08/16/2005
• Est. Priority Date: 08/27/2003
• Status: Active Grant

• Current Assignee: Datex-Ohmeda Incorporated (General Electric Company)
• Original Assignee: Datex-Ohmeda Incorporated (General Electric Company)
• Inventors: Terry, Alvin Mark
• Primary Examiner(s): WINAKUR, ERIC FRANK
• Application Number:
  US10/927,837
• Publication Number:
  US 20050049470A1
• Time in Patent Office: 354 Days
• Field of Search: 600/310, 600/322, 600/323, 600/336, 600/481, 600/483, 600/485, 600/500, 600/502, 128/925
• US Class Current: 600/336
• CPC Class Codes: A61B 5/14552 Details of sensors specially adapted therefor
• CPC Class Codes: A61B 5/7207 of noise induced by motion artifacts
• CPC Class Codes: A61B 5/7257 using Fourier transforms
• CPC Class Codes: A61B 5/726 using Wavelet transforms
• 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: G16H 50/20 for computer-aided diagnosis, e.g. based on medical expert systems
• CPC Class Codes: Y10S 128/925 Neural network

Abstract

Pulse oximetry is improved through classification of plethysmographic signals by processing the plethysmographic signals using a neural network that receives input coefficients from multiple signal domains including, for example, spectral, bispectral, cepstral and Wavelet filtered signal domains. In one embodiment, a plethysmographic signal obtained from a patient is transformed (240) from a first domain to a plurality of different signal domains (242, 243, 244, 245) to obtain a corresponding plurality of transformed plethysmographic signals. A plurality of sets of coefficients derived from the transformed plethysmographic signals are selected and directed to an input layer (251) of a neural network (250). The plethysmographic signal is classified by an output layer (253) of the neural network (250) that is connected to the input layer (251) by one or more hidden layers (252).

Claims

1. A pulse oximeter comprising:

a first optical signal source operable to emit an optical signal characterized by a first wavelength;

a second optical signal source operable to emit an optical signal characterized by a second wavelength different than said first wavelength;

a detector operable to receive said first and second optical signals after said first and second optical signals are attenuated by a patient tissue site of a patient, said detector being further operable to provide a detector output signal representative of said attenuated first and second optical signals; and

a processor enabled to obtain first and second time domain plethysmographic signals from the detector output signal and classify at least one of the first and second time domain plethysmographic signals using a neural network, said neural network receiving input coefficients derived from at least one transform of said at least one of said first and second time domain plethysmographic signals.

2. The pulse oximeter of claim 1 wherein said at least one of said first and second time domain plethysmographic signals is classified by said neural network as being associated with at least one of a plurality of different types of predetermined signal conditions.

3. The pulse oximeter of claim 2 wherein the plurality of different types of predetermined signal conditions comprise:

(1) Normal adult plethysmographic signal;

(2) Normal baby plethysmographic signal;

(3) Fixed frequency tapping motion without plethysmographic signal;

(4) Fixed frequency tapping motion with plethysmographic signal;

(5) Moving frequency tapping motion without plethysmographic signal;

(6) Moving frequency tapping motion with plethysmographic signal;

(7) Clenching motion without plethysmographic signal;

(8) Clenching motion with plethysmographic signal;

(9) Baby-kicking motion without plethysmographic signal;

(10) Baby-kicking motion with plethysmographic signal;

(11) Irregular plethysmographic signal associated with patient arrhythmia;

(12) Low perfusion condition plethysmographic signal;

(13) Plethysmographic signal with fast breathing;

(14) Plethysmographic signal with shallow breathing;

(15) Plethysmographic signal with extended breath holding; and

(16) No plethysmographic signal.

4. The pulse oximeter of claim 2 wherein said neural network is trained prior to field use with data sets representative of each of said plurality of different types of predetermined signal conditions.

5. The pulse oximeter of claim 4 wherein said neural network comprises a fuzzy neural network that has been trained prior to field use in a manner achieving fuzzification of said pre-trained neural network.

6. The pulse oximeter of claim 1 wherein said at least one transform comprises at least one of a spectral transform, a bispectral transform, a cepstral transform, and Wavelet filter bank transform.

7. The pulse oximeter of claim 6 wherein, in addition to said coefficients, said neural network further receives at least one signal characteristic derived from at least one of the first time domain plethysmographic signal, the second time domain plethysmographic signal, and said at least one transform of said at least one of said first and second time domain plethysmographic signals.

8. The pulse oximeter of claim 7 wherein said at least one signal characteristic comprises at least one of:

(1) An RMS energy measure;

(2) A spikiness measure;

(3) A spectral jitter measure;

(4) A spectral shimmer measure;

(5) A spectral smear measure;

(6) A cepstral peak jitter measure; and

(7) A cepstral peak position measure.

9. The pulse oximeter of claim 1 wherein said neural network comprises one of a feed-forward network and a recursive network.

10. The pulse oximeter of claim 9 wherein inputs to an intermediate layer of said neural network are used to fuzzify the input coefficients to an initial layer of said neural network.

11. The pulse oximeter of claim 1 wherein said first wavelength is within the range of infrared light wavelengths and said second wavelength is within the range of red light wavelengths.

12. The pulse oximeter of claim 1 wherein said pulse oximeter further comprises:

a drive system operable to cause operation of said first and second optical signal sources such that each of said first and second optical signal sources emit first and second optical signals, respectively, in accordance with a multiplexing method;

a sampler operable to sample the detector output signal at a desired sampling rate and output a signal having a series of sample values representative of said attenuated first and second optical signals; and

wherein said processor comprises a digital processor, said digital processor being further operable to demultiplex the series of sample values to obtain said first and second time domain plethysmographic signals.

13. The pulse oximeter of claim 12 wherein said desired sampling rate is at least 50 Hz.

14. The pulse oximeter of claim 12 wherein said multiplexing method comprises at least one of frequency division multiplexing, time division multiplexing, and code division multiplexing.

15. A method of processing a plethysmographic signal obtained from a patient, the plethysmographic signal being obtained in a first signal domain, said method comprising the steps of:

transforming the plethysmographic signal from the first domain to a plurality of signal domains different from the first domain to obtain a corresponding plurality of transformed plethysmographic signals, each transformed plethysmographic signal being in one of the different signal domains;

selecting a plurality of sets of coefficients, each set of coefficients being derived from a corresponding one of the transformed plethysmographic signals;

inputting the sets of coefficients to a neural network; and

classifying the plethysmographic signal based on an output from the neural network.

16. The method of claim 15 wherein the first domain comprises the time domain and wherein said step of transforming comprises:

transforming the plethysmographic signal from the time domain to the spectral domain to obtain a spectral domain plethysmographic signal; and

transforming the spectral domain plethysmographic signal to the cepstral domain to obtain a cepstral domain plethysmographic signal.

17. The method of claim 16 wherein said step of transforming the plethysmographic signal from the time domain to the spectral domain comprises:

performing a Fourier transformation on the time domain plethysmographic signal.

18. The method of claim 16 wherein said step of transforming the plethysmographic signal from the spectral domain to the cepstral domain comprises:

performing a Fourier transformation on the spectral domain plethysmographic signal.

19. The method of claim 16 wherein said step of transforming further comprises:

transforming the plethysmographic signal from the time domain to the bispectral domain to obtain a bispectral domain plethysmographic signal.

20. The method of claim 19 wherein said step of transforming the plethysmographic signal from the time domain to the bispectral domain is performed in accordance with the following expression:

B(ω

₁,ω

₂)=E[F(ω

₁)F(ω

₂){overscore (F(ω

₁+ω

₂))}]

where ω

₁ and ω

₂ are the frequencies present in spectrums of the time domain plethysmographic signal.

21. The method of claim 16 wherein said step of transforming further comprises:

applying a Wavelet filter bank transform to the time domain plethysmographic signal to obtain a Wavelet filtered domain plethysmographic signal.

22. The method of claim 15 further comprising:

deriving at least one signal characteristic using information included in at least one of the first domain plethysmographic signal or the transformed plethysmographic signals; and

inputting the at least one signal characteristic to the neural network in addition to the sets of coefficients.

23. The method of claim 22 wherein, in said step of deriving, the at least one signal characteristic comprises at least one of:

(1) An RMS energy measure;

(2) A spikiness measure;

(3) A spectral jitter measure;

(4) A spectral shimmer measure;

(5) A spectral smear measure;

(6) A cepstral peak jitter measure; and

(7) A cepstral peak position measure.

24. The method of claim 15 wherein, in said step of selecting, each set of coefficients corresponds with a plurality of frames of its corresponding transformed plethysmographic signal.

25. The method of claim 15 wherein said step of classifying comprises:

associating the output from the neural network with at least one of a plurality of different types of predetermined signal conditions.

26. The method of claim 25 wherein, in said step of associating, the plurality of different types of predetermined signal conditions comprise:

(1) Normal adult plethysmographic signal;

(2) Normal baby plethysmographic signal;

(3) Fixed frequency tapping motion without plethysmographic signal;

(4) Fixed frequency tapping motion with plethysmographic signal;

(5) Moving frequency tapping motion without plethysmographic signal;

(6) Moving frequency tapping motion with plethysmographic signal;

(7) Clenching motion without plethysmographic signal;

(8) Clenching motion with plethysmographic signal;

(9) Baby-kicking motion without plethysmographic signal;

(10) Baby-kicking motion with plethysmographic signal;

(11) Irregular plethysmographic signal associated with patient arrhythmia;

(12) Low perfusion condition plethysmographic signal;

(13) Plethysmographic signal with fast breathing;

(14) Plethysmographic signal with shallow breathing;

(15) Plethysmographic signal with extended breath holding; and

(16) No plethysmographic signal.

27. The method of claim 25 further comprising:

training the neural network prior to field use with data sets representative of each of said plurality of different types of predetermined signal conditions.

28. The method of claim 27 wherein, in said step of training, the neural network is trained prior to field use in a manner achieving fuzzification of the neural network.

29. The method of claim 27 wherein, in said step of training, the neural network is trained in accordance with one of a backpropagation learning procedure and a simulated annealing learning procedure.

30. The method of claim 29 wherein, in said step of training, the backpropagation learning procedure is implemented with fuzzy logic control.

31. The method of claim 15 further comprising:

transmitting an optical signal through a tissue site of the patient to obtain the first domain plethysmographic signal.

32. A method of training a neural network to classify a plethysmographic signal obtained from a patient, said method comprising the steps of:

selecting a plurality of first domain plethysmographic signal data sets associated with a plurality of different types of predetermined signal conditions from a database of plethysmographic signal data sets;

transforming the first domain plethysmographic signal data sets to other signal domains different than the first domain to obtain a corresponding plurality of transformed plethysmographic signal data sets;

extracting a plurality of sets of coefficients from the transformed plethysmographic signal data sets, each set of coefficients being extracted from a corresponding one of the transformed plethysmographic signal data sets;

using the sets of extracted coefficients as inputs to the neural network; and

adjusting weighting values associated with connections between neurons in the neural network in accordance with a learning procedure.

33. The method of claim 32 wherein, in said step of selecting, the plurality of different types of predetermined signal conditions comprise:

(1) Normal adult plethysmographic signal;

(2) Normal baby plethysmographic signal;

(3) Fixed frequency tapping motion without plethysmographic signal;

(4) Fixed frequency tapping motion with plethysmographic signal;

(5) Moving frequency tapping motion without plethysmographic signal;

(6) Moving frequency tapping motion with plethysmographic signal;

(7) Clenching motion without plethysmographic signal;

(8) Clenching motion with plethysmographic signal;

(9) Baby-kicking motion without plethysmographic signal;

(10) Baby-kicking motion with plethysmographic signal;

(11) Irregular plethysmographic signal associated with patient arrhythmia;

(12) Low perfusion condition plethysmographic signal;

(13) Plethysmographic signal with fast breathing;

(14) Plethysmographic signal with shallow breathing;

(15) Plethysmographic signal with extended breath holding; and

(16) No plethysmographic signal.

34. The method of claim 32 wherein, in said step of transforming the first domain plethysmographic signal data sets to other signal domains, the signal domains different than the first domain comprise spectral, cepstral, bispectral and Wavelet filter bank signal domains.

35. The method of claim 32 wherein, in said step of adjusting, the weighting values are adjusted in accordance with one of a backpropagation learning procedure and a simulated annealing learning procedure.

36. The method of claim 32 further comprising:

deriving at least one signal characteristic using information included in at least one of the first domain plethysmographic signal data sets and the transformed plethysmographic signal data sets; and

using the at least one signal characteristic as an input to the neural network in addition to the extracted coefficients.

37. The method of claim 36 wherein, in said step of deriving, the at least one signal characteristic comprises at least one of:

(1) An RMS energy measure;

(2) A spikiness measure;

(3) A spectral jitter measure;

(4) A spectral shimmer measure;

(5) A spectral smear measure;

(6) A cepstral peak jitter measure; and

(7) A cepstral peak position measure.

38. The method of claim 32 further comprising:

storing plethysmographic signals of interest.

39. The method of claim 38 wherein, in said step of storing, the plethysmographic signal of interest is stored in a memory device of a pulse oximeter.

40. The method of claim 38 further comprising:

retrieving the stored plethysmographic signal of interest; and

adding the plethysmographic signal of interest to the database of plethysmographic signal data sets.

41. A method of providing information relating to a physiological condition of a patient based on at least one plethysmographic signal obtained from the patient, the plethysmographic signal being obtained in a first signal domain, said method comprising the steps of:

transforming the plethysmographic signal from the first domain to a plurality of signal domains different from the first domain to obtain a corresponding plurality of transformed plethysmographic signals, each transformed plethysmographic signal being in one of the different signal domains;

classifying the plethysmographic signal based on an output from a neural network, wherein the output of the neural network is based on input coefficients derived from at least one of the transformed plethysmographic signals; and

selecting a technique for determining the physiological condition of the patient based on the classification.

42. The method of claim 41 wherein the physiological condition of the patient comprises a pulse rate.

43. The method of claim 41 wherein at least two plethysmographic signals corresponding to different optical wavelengths are transformed and classified, and wherein the physiological condition of the patient comprises an SpO2 value.

44. The method of claim 41 wherein at least two plethysmographic signals corresponding to different optical wavelengths are transformed and classified, the plethysmographic signals being transformed using at least a Wavelet filter bank transform, and wherein the physiological condition of the patient comprises a respiration index.

45. The method of claim 41 wherein the first domain comprise the time domain and the plurality of signal domains different than the first domain comprise spectral, cepstral, bispectral and Wavelet filtered domains.