Heartbeat disease diagnosis using text-based approaches

• 第一著者: Khorasani, Ehsan Safar
• フォーマット: 学位論文
• 言語: 英語; 英語
• 出版事項: 2011
• 主題: Heart - Diseases - Diagnosis - Programmed instruction; Heart - Sounds - Programmed instruction; Heart - Examination - Programmed instruction
• オンライン･アクセス: http://psasir.upm.edu.my/id/eprint/27375/1/FSKTM%202011%2014R.pdf

Heart sound signals are the important asset for heart examination in primary healthcare centers to aid significantly in the diagnosis of heart diseases. Interpretation of heart sounds is a problematic and difficult skill that requires cardiology specialists. The diagnosis of heart disease from heart sound can differ between cardiologists and would require more detailed and expensive tests. However, heart disease diagnosis by heartbeat is preferable and still widely used as the first step to diagnosis. Computer aided auscultation has emerged as a costeffective technique to analyze and interpret the heart sounds. Digital heart sound recordings with background noise, similarity among heart diseases, recording environment conditions, auscultation body points makes detection of heart diseases complicated. There are several methods for automated detection and classification of heart diseases and heart sound analysis that have been proposed. Some of them used Artificial Neural Network method for detection and classification of heart sounds. Another technique that it used for diagnosis the heart problem is Hidden Markov Model (HMM) that they suggest HMM for segmentation of heart sound recorded for clinical and classification purpose. However, to the best knowledge of the researcher, no prior study has encoded heart sound to text string. In this study, we propose a feasible technique for developing a heartbeat sound retrieval system using text encoding techniques which is useful towards automated heart disease detection. The audio format of heart sound recordings are preprocessed and transcribed into the MIDI format. The MIDI files are then encoded to text strings using the pitch and duration information based on n-gram, these text strings then form musical words. These text strings are then indexed and tested for retrieval using both database and Information Retrieval (IR) systems. The Longest Common Subsequence (LCS) matching algorithm was used for identifying similarities from the database. With IR, full text indexing of the recordings was used and retrieved using known item searches from a search engine. The feasibility of these text encoding techniques were shown from retrieval experiments with around 100 digital heart sound recordings. Overall, experimental results performed clearly showed the feasibility of using proposed text encoding techniques for diagnosing heart problems. Thus, it can be said that the results presented for heart sound retrieval system are very promising for queries in Normal and Abnormal heart sound categories.