Method and apparatus for processing patient sounds

1. A method of operating a computational device to process patient sounds for diagnosing a respiratory disease, the method comprising:

having a sound receiving device configured to receive the patient sounds from a patient and form a corresponding digitized sound signal;

computing feature vectors for each of a plurality of sub-blocks of the digitized sound signal including two or more of: Mel-frequency cepstral coefficients (MFCCs), entropy features, Zero Crossing Rate, and Non-Gaussianity;

operating a trained cough detection pattern classifier to respond to the feature vectors to produce a cough detection pattern classifier output signal;

filtering said output signal to produce a smoothed output signal;

thresholding the smoothed output signal to identify candidate cough segments in the digitized sound signal;

classifying candidate cough segments to be cough segments based upon their duration being longer than a minimum cough length and shorter than a maximum cough length;

forming a plurality of sub-segments of each of the cough segments and computing feature vectors for each of the sub-segments, these feature vectors including two or more of: Mel-frequency cepstral coefficients (MFCCs), entropy features, Zero Crossing Rate, and Non-Gaussianity features;

applying said feature vectors of the plurality of sub-segments to a trained diseased cough pattern classifier to classify each of the cough segments as being one of diseased and non-diseased for a particular respiratory disease;

computing a diseased cough index indicating a proportion of cough segments classified as diseased relative to all of the cough segments;

deeming the patient to be suffering from the particular respiratory disease based on the diseased cough index; and

presenting a diagnosis of the particular respiratory disease on a display under control of the computational device.

2. A method according to claim 1 , wherein the forming a plurality of sub-segments of the cough segments comprises forming three sub-segments.

3. A computational device including at least one electronic processor in communication with an electronic memory containing instructions for the processor to carry out a method according to claim 1 .

4. A computational device according to claim 3 , wherein the computational device comprises a mobile telephone that is programmed to carry out said method.

5. A method according to claim 1 , wherein the computing feature vectors for each of a plurality of sub-blocks of the digitized sound signal includes extracting any one of the following groups of features:

MFCCs and Formant Frequency and Zero Crossing Rate and Shannon Entropy and Non-Gaussianity features; or

Zero Crossing Rate and Shannon Entropy and Non-Gaussianity features; or

MFCCs and Zero Crossing Rate and Formant frequency and Non-Gaussianity features.

6. A method according to claim 1 , wherein the trained cough detection pattern classifier comprises an artificial neural network.

7. A method according to claim 6 , wherein the artificial neural network comprises a time delay neural network (TDNN).

8. A method according to claim 7 , wherein the TDNN has a hidden layer between an output layer and an input layer with 10 to 50 neurons in the hidden layer.

9. A method according to claim 1 , further comprising applying higher order statistical analysis to the cough segments for further classification of the cough segments.

10. A method according to claim 9 , wherein the higher order statistical analysis comprises calculating a bispectrum.

11. A method according to claim 1 , wherein the forming a plurality of sub-segments of each of the cough segments and computing feature vectors for each of the sub-segments further comprises:

determining one or more of: Bispectrum Index (BSG),

Formants Frequencies (FF), Log energy (Log E), and Kurtosis (Kurt) features for each of the cough sub-segments.

12. A method according to claim 11 , wherein the trained diseased cough pattern classifier comprises any one of: a logistic regression model; an artificial neural network; a Bayes classifier; a hidden Markov model; and a support vector machine.

13. A method according to claim 12 ,

wherein the trained diseased cough pattern classifier has been trained with a training set that includes non-pneumonic sounds recorded from patients suffering from one or more of the following complaints: Asthma, Bronchitis, Rhinopharyngitis, wheezing, tonsilopharanzitis, heart disease, larangomalaysia, malaria, and foreign body inhalation.

14. A method according to claim 1 , wherein the trained diseased cough pattern classifier comprises any one of: a logistic regression model; an artificial neural network; a Bayes classifier; a hidden Markov model; and a support vector machine.

15. A method according to claim 14 ,

wherein the trained diseased cough pattern classifier has been trained with a training set that includes non-pneumonic sounds recorded from patients suffering from one or more of the following complaints: Asthma, Bronchitis, Rhinopharyngitis, wheezing, tonsilopharanzitis, heart disease, larangomalaysia, malaria, and foreign body inhalation.

16. A method according to claim 1 , wherein the particular respiratory disease comprises pneumonia.

17. A method according to claim 16 , wherein the diseased cough index is a pneumonic cough index.

18. A method according to claim 1 , wherein the presenting of the diagnosis of the disease related state on the display comprises presenting a diagnosis of pneumonia.

19. A method according to claim 1 , wherein the patient sounds are acquired with one or more microphones that are removed from physical contact with the patient.