Abstract:
Heart valves, responsible for correct pumping of blood to entire body generate certain
sounds during its functionality called as heart sounds. Listening and interpretation of
these sounds using stethoscope is known as auscultation. Heart sounds commonly known
as phonocardiogram signal provide valuable information for correct identification of any
heart disease, if interpreted correctly. These sounds though audible but need an extensive
practice and skills to be correctly understood. Any illness of heart valves like murmurs
though appear in these sounds but are very difficult to be correctly identified by the
cardiologist. These murmurs are even further having many types. Phonocardiogram
signals can be utilized more efficiently by the cardiologists and medical officers when
they are converted into some easily interpretable form rather through a conventional
stethoscope. This research work is carried out with an aim to segment the heart sounds by
identify the correct locations of first and second heart sounds and classify them to
identify any illness of heart valves. Correct identification of S1 and S2 is an important but
less addressed issue of segmentation problem. By correctly segmenting the
phonocardiogram signal into its subparts any illness can easily be isolated, detected and
classified. To undertake the segmentation task I have used the effectiveness of k-means
clustering which is used to segregate and label the heart sounds as S1 and S2 sounds. For
correct classification of illness I have used a novel feature set by combing the temporal
and frequency domain characteristics. All distinct features from both the domains are
made part of feature vector for classification purpose. To test the effectiveness of my
method, I used PASCAL Classifying Heart Sounds Challenge 2011(PASCALCHSC2011) dataset and successfully obtained improved results for segmentation,
identification and classification problem than any of the challenge participants.
