張璞曾臺灣大學：電機工程學研究所張峰嘉Chang, Feng-ChiaFeng-ChiaChang2007-11-262018-07-062007-11-262018-07-062004http://ntur.lib.ntu.edu.tw//handle/246246/52943N/AAbstract This thesis focuses on two important objectives. Firstly, construct an easy, low-cost and high performance respiratory sound capturing system in accordance with CORSA standard. Then, develop a high performance record-position-independent and system-easy-to-transplant wheeze episode detecting algorithm based on spectrogram image processing of respiratory sounds. The analog circuit design process is followed by direct PC soundcard digitalization for sound capturing. Spectrogram images generated from recorded sounds are passed through a 2D bilateral filter to do edge-preserving smoothing. Then the processed data go through edge detecting procedures, wheeze-with or -without sounds are elaborately recognized. The result shows both high sensibility of 0.946 and specificity of 0.92 in qualitative analysis of wheezes from different recording locations. Due to its high efficiency, great performance and easy-to-implement characteristics, 2-D bilateral filtering can be perfectly adapted for real-time monitoring of respiratory sounds for wheezes detection.Contents Contents Ⅰ List of Figures Ⅴ List of Tables Ⅵ Abstract Ⅶ 1 Introduction 1.1 Background on Auscultation 1 1.2 Introduction of Respiratory Sounds 1.2.1 Normal Breath Sounds 4 1.2.1.1 Normal Lung Sounds 5 1.2.1.2 Normal Tracheal Sounds 7 1.2.2 Abnormal Breath Sound - Wheezes 1.2.2.1 Continuous Adventitious Sound 10 1.2.2.2 Production Mechanism 11 1.2.2.3 Classification 11 1.2.2.4 Characteristics 12 1.2.2.5 Parametric Quantification 14 1.3 Motivations 14 2 System Architecture 2.1 Capturing and Preprocessing of Respiratory Sounds 2.1.1 CORSA Standard for Analog Acquisition Chain 17 2.1.1.1 Sensors 18 2.1.1.2 Pre-amplifier and Final-amplifier 20 2.1.1.3 Filters 21 2.1.1.4 Noise and Interference 24 2.1.1.5 Electrical Safety 25 2.1.2 Respiratory Sound Capturing System – Analog Part 25 2.1.2.1 Sensor 28 2.1.2.2 Pre-amplifier 30 2.1.2.3 Filters 31 2.1.2.4 Electrical Safety 32 2.1.3 Respiratory Sound Capturing System – Digitization 32 2.1.4 System Verification 2.1.4.1 Clipping and Overload 34 2.1.4.2 Frequency Response of Analog Pre-filters 34 2.1.4.3 Frequency Response of Soundcard 34 2.1.4.4 Noise and Interference 35 3 Algorithm for Wheeze Episode Detection 3.1 Characteristics of Wheeze Episode in Spectrogram 37 3.2 Edge Reserving Filter – Bilateral Filter 3.2.1 Background 41 3.2.2 General Idea 41 3.2.3 Filter Parameters 43 3.3 Wheezes Detecting Algorithm based on 2D Bilateral Filtering of Spectrogram Images 3.3.1 Existing Algorithms for Wheeze Detection 45 3.3.2 Ultimate Goal of Detection 47 3.3.3 Algorithm Implementation 48 4 Result and Discussion 4.1 Algorithm Validation 4.1.1 Subjects 55 4.1.2 Performance Analysis 56 4.2 Discussion 4.2.1 Error Detection 57 5 Conclusion and Future Work 61 Reference List of Figures 1-1 The Father of Auscultation --- Laennec 1 1-2 Sites for Auscultation 2 1-3 Normal Respiratory Sound Patterns 4 1-4 An Example of Lung Sound 6 1-5 An Example of Trachea Sound 8 1-6 Abnormal Respiratory Sound Patterns – Asthma with Wheeze 9 1-7 An Example of Trachea Wheezing Sound 10 2-1 Typical Analog Circuit Configuration 17 2-2 Respiratory Sound Capturing System 25 2-3 Analog Circuit Configuration 26 2-4 Sound Capturing Configuration 28 2-5 Schematic Diagram of KEC – 2738 29 2-6 Visual Inspection of the System Noise and Interference 36 3-1 An Segment of Inspiration and Expiration Phase( IE cycle) of Trachea Wheezing Sound 39 3-2 Comparison between Bilateral Filter and Other Iterative Filters 44 3-3 Block Diagram of Proposed Algorithm 50 3-4 Algorithm Procedures 51 3-5 Same as (a) and (f) in Figure 3-4 with larger scale 53 List of Tables 2-1 Recommendation of Sensor Specifications 18 2-2 Comparison of Microphones in Respiratory Sound Recording 19 2-3 Three Main Parameters of an Amplifier 20 2-4 Descriptions of Pre-amplifier and Final-amplifier 21 2-5 Bandwidths Recommended for Breath Sounds Analysis 22 2-6 Description of Filter Specifications 23 2-7 Typical Noise and Interference in Respiratory Sound Recording 24 2-8 Specifications of the Sensor Used 29 2-9 Specifications of the Pre-amplifier Used 30 2-10 Specifications of the Filter Used 32 2-11 Digitization of Sound 33 3-1 Characteristics of Wheeze 40 3-2 Summary of Comparison 45 3-3 Existing Wheeze Detecting Algorithms 45 3-4 Spectrogram Settings 48 3-5 Parameter Settings of 2D Bilateral Filter 49 4-1 Groups of Respiratory Sound 55 4-2 Results for Separate Respiratory Cycle Analysis 56 4-3 Performance Analysis of the Algorithm 57 4-4 Problems of Error Detection 57919857 bytesapplication/pdfen-USN/Awheezespectrogrambilateral filterCORSArespiratory soundthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/52943/1/ntu-93-R90921030-1.pdf