Interpretation of heart sounds is one of the first steps towards diagnosing a heat disease .The current scenario diagnosis of heart diseases requires special skills and experience along with expensive test which may not be suitable for everyone .Computer aided methods have therefore provided a cost effective method to analyse and interpret heart sounds .It provides the doctor with a tool to make a correct treatment decisions .
There are currently many methods for detection and classification of heart diseases some of which utilise neural networks ,but the following article proposes sound retrieval systems using text based approaches .The proposed approach utilises recording of heart beat sound in MIDI Format.
Advantages of Recording in MIDI format:
MIDI is a popular format used in music productions and it also provides the advantages of a relatively less size as compared to other file formats .MIDI formats works by extracting characteristics of sound like pitch ,volume ,frequency and connecting them into a sequence of notes .The knowledge of heart sounds being in the range of 25-50Hz helps us filter from other sounds as well.Any MIDI converter can serve the purpose of converting audio files collected from heart beats to MIDI format.
Sound Data preprocessing:
In order to carry out sound retrieval using text based format it is important to preprocess the sound collected from different heart beats .Background undesirable noises such as breathing sound or sound of stethoscope touch cause much problem in data collection.For preprocessing the audio it is first converted into a wave format where it amplifies the heart recordings and then noise reduction techniques are applied on it.
Image before and after preprocessing
Summarised steps of the proposed method:
1.The audio format of the heart sound recordings are processed and converted into a MIDI Format .
2.The MIDI format files are then encoded into text strings using N-grams , the method of converting them into N grams is given below .
3.These strings are then indexed and stored into Information Retrieval systems .
4.In order to identify similarities with the database Longest Common Subsequence(LCS) technique is used.
To obtain text strings from MIDI encodings ,the N-gram approach will be used .The N-grams technique of encoding as musical words have been used for other methods as well for indexing ,searching and retrieval data from musical notes .By adopting this scheme heartbeat sounds can be converted into musical notation.
Steps to obtain N gram sequences :
1.Divide the MIDI file format data using a gliding window approach into overlapping windows of n different sizes.
2.Obtain all possible combinations of strings from each and every window .
N-grams are constructed from the interval sequences within each window .Intervals serve as a great method for finding patterns in sound .For a sequence of pitches an interval sequencing is obtained as :
Another term rhythmic ratio sequence is obtained for sequences with n different onset times .
The N-grams are constructed are in patterns of interval followed by rhythmic ratio ,the following formula summaries the n-gram constructed pattern.
String matching Method:
In order to measure the similarity of searched query and records in the database ,Longest Common subsequence (LCS)technique is used ,algorithm which examines all the words in every stored text files to match the search words supplied by the heartbeat sound.
References:
2. T Topal, H Polat, I Guler, "Software Development For the Analysis of Heartbeat Sounds with LabView in Diagnosis of Cardiovascular Disease", Journal of Medical System, vol. 32, pp. 409-21, 2008.
Comments
Post a Comment