Research On The Visual Recognition For Adventitious Respiratory Sounds

Adventitious respiratory sounds are non-periodic transient signals,its instant identification and analysis is a necessary means to monitor the status of the respiratory system and to improve the treatment.Adventitious respiratory sounds are caused by complicated reasons,occurred in different locations and collected by different methods in clinical practice.Therefore,lung sounds are more difficult to be analyzed and identified than heart sounds.Most current researches of lung sounds are relying on the methods of frequency and numerical analysis.Due to the lack of visual basis and the limited samples in clinic,the recognized methods for automatic classifications of lung sounds are difficult to be recognized and promoted.In this thesis,a visual recognition method of adventitious respiratory sounds is proposed to provide clinicians intuitive classification basis and effective identification methods of lung sounds.The following research work is carried out in this thesis:(1)A spectral image model of typical adventitious respiratory sounds is proposed,and a new method for the recognition of lung sounds based on this model is designed.The innovations of the new model and method are shown as follows.First of all,previous studies on the two typical adventitious respiratory sounds,crackles(also known as wet rales)and wheezes(also known as dry rales),only describe their spectra but no characteristics analyses.So the new model describes the spectra features of crackles and wheezes,which could be the basis of visual recognition of these lung sounds.Secondly,the new model combines the spectra characteristics of two typical adventitious respiratory sounds and forms a unified mathematical model and characteristic calculation method.Different lung sounds have different numerical characters in this model,so the model could be the numerical analysis foundation for visual recognition method of adventitious respiratory sounds.(2)According to the unified spectral image model of adventitious respiratory sounds,the visualization recognition methods of crackles are proposed.These methods are the VC-STFT algorithm based on STFT transform,the VCWT algorithm based on wavelet transform and the VCST algorithm based on S transform.Then the data analyses were carried out by these specific algorithms.The validity of the proposed visual recognition methods of crackles are verified by the experimental analyses.(3)According to the unified spectral image model of adventitious respiratory sounds,the visual recognition methods of the wheezes are proposed.They include the VWHT algorithm based on Hough transform,the VW-STFT algorithm based on STFT transform,the VWWT algorithm based on wavelet transform and the VWST algorithm based on S transform.The specific algorithms are designed and the data analyses are carried out.The validity of the visual recognition methods of wheezes are verified by the experimental analyses.(4)According to the fusion spectral image model of adventitious respiratory sounds,a fusion method is proposed to realize the visual recognition of adventitious respiratory sounds.The specific algorithm name is the VLFR fusion recognition algorithm.The validity and advantage of this method is also verified by the experiments and the data analyses.(5)In order to standardizing the processing of lung sounds signals,a unified treatment method is proposed for the clinical and international sharing lung sounds data.These lung sounds files were unified processed with this method and stored in the experimental lung sound library.This method provides the database for the processing of lung sounds which were collected from different ways.Based on this work,the above spectral characteristic model of adventitious respiratory sounds could be summed up.