System and method for classifying a heart sound

• US 9,168,018 B2
• Filed: 02/18/2015
• Issued: 10/27/2015
• Est. Priority Date: 08/25/2010
• Status: Active Grant

• Current Assignee: Diacoustic Medical Devices (Pty) Ltd.
• Original Assignee: Diacoustic Medical Devices (Pty) Ltd.
• Inventors: Pretorius, Eugene, Strydom, Otto Marsl, Cronje, Matthys Louwrens
• Primary Examiner(s): Gedeon, Brian T
• Application Number:
  US14/624,960
• Publication Number:
  US 20150164466A1
• Time in Patent Office: 251 Days
• Field of Search: None
• US Class Current: 1/1
• CPC Class Codes: A61B 5/02 Detecting, measuring or recording pulse, heart rate, blood pressure or blood flow; Combined pulse/heart-rate/blood pressure determination; Evaluating a cardiovascular condition not otherwise provided for, e.g. using combinations of techniques provided for in this group with electrocardiography or electroauscultation; Heart catheters for measuring blood pressure
• CPC Class Codes: A61B 5/349 Detecting specific parameters of the electrocardiograph cycle
• CPC Class Codes: A61B 5/352 Detecting R peaks, e.g. for synchronising diagnostic apparatus; Estimating R-R interval
• 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: A61B 7/04 Electric stethoscopes microphones, acoustic transducers therefor H04R

Abstract

A method and system for electronically classifying a pre-processed heart sound signal of a patient as functional (normal) or pathological is provided. The pre-processed patient heart sound signal is segmentised and features are extracted therefrom (104) to build up a feature vector which is representative of the heart sound signal. The feature vector is then fed to a diagnostic decision support network (105) comprising a plurality of artificial neural networks, each relating to a known heart pathology, which is in turn used to conduct the classification.

Claims

1. A computer-implemented method of classifying, a recorded, digitized acoustic heart sound signal of a patient, the heart sound signal having been pre-processed so as to be parsed into respective heart beat cycles with S1 and S2 heart sounds identified, the method being performed by a computing device and including the steps of:

by a computing device, segmenting and extracting features from heart beat cycles making up the heart sound signal and compiling the extracted features into a feature vector, the extracted features including at least energy features extracted by utilizing a power spectrum calculation conducted on the heart beat cycles of the heart sound signal, and frequency domain features extracted by utilizing a frequency domain transformation of the heart beat cycles of the heart sound signal;

feeding, by a computing device, the feature vector into a network of a plurality of artificial neural networks, each artificial neural network being associated with a specific pathology and being trained with at least features extracted from similarly pre-processed heart sound signals associated with the specific pathology;

determining, by a computing device, whether the heart sound signal exemplifies features corresponding to the artificial neural network associated with any one or more of the pathologies; and

classifying, by a computing device, the heart sound signal as normal or containing a murmur associated with one or more of the pathologies according to the determination.

2. A method as claimed in claim 1, wherein determining whether the heart sound signal exemplifies features corresponding to the artificial neural network associated with any one or more of the pathologies includes determining, by a computing device, a confidence measure to indicate the confidence with which a diagnosis is made.

3. A method as claimed in claim 1, wherein extracting features from heart beat cycles making up the heart sound signal includes calculating, by a computing device, the Shannon envelope of the heart beat cycles using S1 and S2;

extracting, by a computing device, eight Shannon envelope features from the Shannon envelope, six of which are located in the systole region of a heart beat cycle and two in the S2 region;

calculating, by a computing device, the Shannon envelope'"'"'s mean values in the regions of the Shannon envelope features; and

normalising them, by a computing device, by scaling the eight Shannon features to have a maximum value of 1.

4. A. method as claimed in claim 1 wherein segmenting and extracting features from heart beat cycles making up the heart sound signal includes segmenting, by a computing device, heart beat cycles, including S1 and 52 sounds and the systole region, into bins;

calculating, by a computing device, a Short-Time Fourier Transform (“

STFT”

) for overlapping bins;

splitting, by a computing device, resulting STFTs into low frequency and high frequency values;

extracting, by a computing device, low frequency and high frequency values from the bin segments;

scaling, by a computing device, low frequency and high frequency values to the same magnitude; and

normalising, by a computing device, the resulting low frequency and high frequency values to the range [−

1;

1].

5. A method as claimed in claim 1, wherein a diagnostic decision support network is provided which comprises the artificial neural networks for each pathology, the diagnostic decision support network being, used to determine if the heart sound signal corresponds to the artificial neural networks of one or more of the pathologies.

6. A method as claimed in claim 1, wherein an output of each artificial neural network is used to calculate, by a computing device, a percentage of heart beat cycles making up the heart beat signal classified as normal and a percentage of heart beat cycles making up the heart beat signal. classified as containing a murmur associated with one or more of the pathologies.

7. A method as claimed in claim 6, wherein the heart sound signal is classified, by a computing device, as normal if at least 60% of the heart beat cycles making up the heart sound signal were classified as normal;

and wherein the heart sound signal is classified, by a computing device, as pathological if more than 40% of the heart beat cycles making up the heart sound signal were classified as containing, a murmur associated with one or more of the pathologies.

8. A system for classifying a recorded. digitized acoustic heart sound signal of a patient, the heart sound signal having been pre-processed so as to be parsed into respective heart beat cycles with S1 and S2 heart sounds identified, the system including at least one processing circuit and a computer-readable non-transitory medium coupled to the processing circuit, the computer-readable medium comprising code executable by the processing, circuit for implementing a method comprising:

receiving, by a computing device, the heart sound signal;

by a computing device, segmenting and extracting, features from heart beat cycles making up the heart sound signal and compiling the extracted features into a feature vector, the extracted features including at least energy features extracted by utilizing a power spectrum calculation conducted on the heart beat cycles of the heart sound signal, and frequency domain features extracted by utilizing, a frequency domain transformation of the heart beat cycles of the heart sound signal;

feeding, by a computing device, the feature vector into a network of a plurality of artificial neural networks, each artificial neural network being associated with a specific pathology and being trained with at least features extracted from similarly pre-processed heart sound signals associated with the specific pathology;

determining, by a computing device, whether the heart sound signal exemplifies features corresponding to the artificial neural network associated with any one or more of the pathologies; and

classifying, by a computing device, the heart sound signal as normal or containing a murmur associated with one or more of the pathologies according to the determination.

9. A system as claimed in claim 8, which includes a diagnostic decision support network comprising the artificial neural networks and wherein the method includes the step of utilising the diagnostic decision support network to provide a diagnosis.

10. A system as claimed in claim 9, wherein the method further includes the step of determining, by a computing device, a confidence measure to indicate the confidence with which a diagnosis is made.

11. A system as claimed in claim 8, wherein the method further includes the steps of calculating, by a computing device, the Shannon envelope of the heart beat cycles using S1 and S2;

extracting, by a computing device, eight Shannon envelope features from the Shannon envelope, six of which are located in the systole region of a heart beat cycle and two in the S2 region;

calculating, by a computing device, the Shannon envelope'"'"'s mean values in the regions of the Shannon envelope features; and

normalising them, by a computing device, by scaling the eight Shannon features to have a maximum value of 1.

12. A system as claimed in claim 8, wherein the method further includes the steps of segmenting, by a computing device, heart beat cycles, including S1 and S2 sounds and the systole region, into bins;

calculating, by a computing device, a Short-Time Fourier Transform (“

STFT”

) for overlapping bins;

splitting, by a computing device, resulting STFTs into low frequency and high frequency values;

extracting, by a computing device, low frequency and high frequency values from the bin segments;

scaling, by a computing device, low frequency and high frequency values to the same magnitude; and

normalising, by a computing device, the resulting low frequency and high frequency values to the range [−

1;

1].

13. A system as claimed in claim 8, wherein the method further includes the step of using an output of each artificial neural network to calculate, by a computing, device, a percentage of heart beat cycles making up the heart beat signal classified as normal and a percentage of heart beat cycles making up the heart beat signal classified as containing, a murmur associated with one or more of the pathologies.

14. A system as claimed in claim 8, wherein the method further includes the steps of classifying the heart sound signal, by a computing device, as normal if at least 60% of the heart beat cycles making up the heart sound signal were classified as normal;

and classifying the heart sound signal, by a computing device, as pathological if more than 40% of the heart beat cycles making up the heart sound signal were classified as containing a murmur associated with one or more of the pathologies.