能連續監測患者症狀達到預防潛在致命反應的穿戴式裝置對於患有慢性阻塞性肺疾病（COPD）和氣喘等呼吸系統疾病的患者可說是一大福音。本論文介紹了名為Stethogram的設備，使用三通道的數位麥克風測量下呼吸道和上呼吸道的聲音。此外，還包括用於監測患者活動的數位IMU，以及用於感測環境二氧化碳及揮發性有機化合物(VOC)的感測器。系統裝置了LTE模組，以進行實時數據上傳。本論文的貢獻包括評估本地處理或儲存選項（包括資料傳輸、機器學習分類和緩存）與在服務器上進行處理之間的比較。評估指標包括功耗、準確性和延遲。實驗結果顯示，Stethogram以低功耗全面監測呼吸系統症狀和環境因素，整合LTE模組實現了實時數據上傳，使醫護人員能夠遠程觀看患者的即時數據。通過整合TinyML技術，Stethogram能夠平衡實時監測和準確性，減少對伺服器持續數據傳輸的需求，從而降低功耗和延遲。

Patients with respiratory diseases such as chronic obstructive pulmonary disease (COPD) and asthma can benefit from wearable devices that continuously monitor the symptoms and conditions that trigger potentially fatal reactions. This thesis describes such a device named Stethogram, which uses a 3-channel digital microphone to measure sounds from lower and upper airways. It also includes a digital inertial measurement unit (IMU) for patient activity and CO2 and VOC for environmental air sensing. The system also includes Long Term Evolution (LTE) module for real-time data uplink. The contributions of this thesis include the evaluation of local processing or storage options including data transmission, machine-learning classification, and buffering compared to on-server processing. Evaluation metrics include power consumption, accuracy, and latency. Experimental results show that Stethogram comprehensively monitors respiratory symptoms and environmental factors in low power consumption. Integrating an LTE module enables real-time data uplink, allowing healthcare professionals to access live patient data remotely. By incorporating TinyML, Stethogram can balance real-time monitoring and accuracy, reducing the need for constant data transmission to the server and reducing power consumption and latency.

Contents i
Acknowledgments v
1 Introduction 1
1.1 Motivation 1
1.2 Contributions 2
1.2.1 System Implementation 2
1.2.2 Data Processing 3
1.3 Thesis Organization 3
2 Background and Related Work 4
2.1 Background: COPD and AECOPD 4
2.1.1 Stages and Groups of Stable COPD 4
2.1.2 AE-COPD 5
2.1.3 Challenges 6
2.2 Related Work 7
2.2.1 AECOPD Prediction Bands 8
2.2.2 Single-Channel Sound Collector 8
2.2.3 Multi-Channel Sound Logger 8
3 Device Design and Data Collection 10
3.1 MCU Architecture 10
3.2 Motion-Tracking Subsystem 11
3.3 Environmental-Sensing Subsystem 11
3.4 Auscultation Subsystem 12
3.5 Power Subsystem 12
3.6 Local Storage Subsystem 13
3.7 Networking Subsystem with Cloud Support 13
3.7.1 LTE 13
3.7.2 TCP/IP Protocol 13
3.7.3 Direct Link Mode 14
3.7.4 Ping-pong Buffer 14
4 CNN Model Training 16
4.1 Datasets 16
4.1.1 Kaggle Database 16
4.1.2 R.A.L.E. Database 17
4.2 COPD Neural Network Classification Model 17
4.3 Evaluation Metrics 18
5 TinyML Model Development 20
5.1 Platform for Training 20
5.2 Model Architecture 20
5.3 Device Deployment 21
5.4 Model Metrics and On-device Performance 22
5.5 Comparative Analysis of Multi-class TinyML Models 23
6 Evaluation 25
6.1 Current-Measuring System 25
6.2 Local Storage Mode vs. Direct Link Mode 25
6.3 Discussion 28
6.4 On-Server CNN Model vs. TinyML Model Performance 28
7 Conclusions and Future Work 30
7.1 Conclusions 30
7.2 Future Work 30
Appendix 37

[ACC¸23] Alvar Agustí, Bartolome R. Celli, Gerard J. Criner, David Halpin, Antonio Anzueto, Peter Barnes, Jean Bourbeau, MeiLan K. Han, Fernando J. Martinez, Maria Montes de Oca, Kevin Mortimer, Alberto Papi, Ian Pavord, Nicolas Roche, Sundeep Salvi, Don D. Sin, Dave Singh, Robert Stockley, M. Victorina López Varela, Jadwiga A.Wedzicha, and Claus F. Vogelmeier. Global initiative for chronic obstructive lung disease 2023 report: Gold executive summary. Archivos de Bronconeumología, 59(4):232–248, 2023.
[Amp13] Amphenol Advanced Sensors. MA100BF103B, Biomedical Chip NTC Thermistors, 7 2013.
[Ard19] Arduino. Arduino Nano 33 BLE. https://docs.arduino.cc/hardware/nano-33-ble, 7 2019.
[AYN¸17] A. Azhari, S. Yoshimoto, T. Nezu, H. Iida, H. Ota, Y. Noda, T. Araki, T. Uemura,
T. Sekitani, and K. Morii. A patch-type wireless forehead pulse oximeter for SpO2
measurement. In 2017 IEEE Biomedical Circuits and Systems Conference (BioCAS),
pages 1–4, 2017.
[CG16] Iuliana Chiuchisan and Oana Geman. Trends in embedded systems for e-Health and biomedical applications. In 2016 International Conference and Exposition on Electrical and Power Engineering (EPE), pages 304–308, 2016.
[Che11] Maheswari Arumugam Chetty. System for detection of vital signals with an embedded system. Afr. J. Inf. Commun. Technol., 6, 2011.
[DGM¸16] James Dieffenderfer, Henry Goodell, Steven Mills, Michael McKnight, Shanshan Yao, Feiyan Lin, Eric Beppler, Brinnae Bent, Bongmook Lee, Veena Misra, Yong Zhu, Omer Oralkan, Jason Strohmaier, John Muth, David Peden, and Alper Bozkurt. Low-power wearable systems for continuous monitoring of environment and health for chronic respiratory disease. IEEE Journal of Biomedical and Health Informatics, 20(5):1251–1264,
2016.
[Edga] Edge Impulse. Edge Impulse, optimize AI for the edge. https://edgeimpulse.com/[2023-08-06].
[Edgb] Edge Impulse. Introducing EON: Neural networks in up to 55% less ram and 35% less ROM. https://www.edgeimpulse.com/blog/introducing-eon.
[FRPC18] Joao T. Fernandes, B. M. Racha, R. P. Paiva, and Tiago J. Cruz. Using low cost embedded systems for respiratory sounds auscultation. In 2018 15th International Symposium on Wireless Communication Systems (ISWCS), pages 1–5, 2018.
[HSI20] Omiya Hassan, Samira Shamsir, and Syed K. Islam. Machine learning based hardware model for a biomedical system for prediction of respiratory failure. In 2020 IEEE International Symposium on Medical Measurements and Applications (MeMeA), pages 1–5, 2020.
[HST¸21] Steven Hicks, Inga Strumke, Vajira Thambawita, Malek Hammou, Pål Halvorsen, Michael Riegler, and Sravanthi Parasa. On evaluation metrics for medical applications of artificial intelligence. 04 2021.
[Ins10] Texas Instruments. INA219 zero-drift, bidirectional current / power monitor with I2C interface. https://www.ti.com/lit/ds/symlink/ina219.pdf, 11 2010.
[KBW¸14] U. Koehler, U. Brandenburg, A. Weissflog, K. Sohrabi, and V. Groß. LEOSound, an innovative procedure for acoustic long-term monitoring of asthma symptoms (wheezing and coughing) in children and adults. Pneumologie (Stuttgart, Germany), 68(4):277—281, April 2014.
[KGK19] Jayanna Kanchikere, A.K. Ghosh, and Kalyankumar Kalyankumar. Embedded patient monitoring system. International Journal of Power Electronics and Drive Systems, 10(1):388, 2019.
[KHJ¸21] Yoonjoo Kim, YunKyong Hyon, Sung Soo Jung, Sunju Lee, Geon Yoo, Chaeuk Chung, and Taeyoung Ha. Respiratory sound classification for crackles, wheezes, and rhonchi in the clinical field using deep learning. Scientific Reports, 11(1):17186, Aug 2021.
[Kno15] Knowles. SPH0645LM4H-B, I2S Output Digital Microphone, 8 2015.
[Lab17] Silicon Labs. EFM32GG11B420F2048GM64-B. https://www.mouser.tw/ProductDetail/Silicon-Labs/EFM32GG11B420F2048GM64-B?qs=gTYE2QTfZfQRUnkc%2FPsgHw%3D%3D, 7 2017.
[MH08] Neil MacIntyre and Yuh Chin Huang. Acute exacerbations and respiratory failure in chronic obstructive pulmonary disease. volume 5, pages 530–535. American Thoracic Society, May 2008.
[MIR¸22] Dennis A. Martilla, Marco C. Iligan, Algerica Raeven R. Ramos, Allan Daraman, Jr., and March Fernan H. Abadines. Wearable tool for breathing pattern recognition and exacerbation monitoring for COPD patients via a device-to-cloud communication model. Journal of Communications, 17(6), 2022.
[R.A] R.A.L.E. Respiratory. The R.A.L.E. Repiratory, hear respiratory sounds. http://www.rale.ca/History.htm.
[RDJ¸18] Joel J. P. C. Rodrigues, Dante Borges De Rezende Segundo, Heres Arantes Junqueira,
Murilo Henrique Sabino, Rafael Maciel Prince, Jalal Al-Muhtadi, and Victor Hugo C.
De Albuquerque. Enabling technologies for the Internet of health things. IEEE Access,
6:13129–13141, 2018.
[RFM¸17] Bruno Rocha, D. Filos, L. Mendes, Ioannis Vogiatzis, Eleni Perantoni, Evangelos Kaimakamis, Pantelis Natsiavas, Ana Oliveira, Cristina Jácome, Alda Marques, Rui Pedro Paiva, Ioanna Chouvarda, P. Carvalho, and N. Maglaveras. A respiratory sound database for the development of automated classification. pages 33–37, 11 2017.
[SCO¸16] Abhilash Sahadevan, Ruth Cusack, B. O’Kelly, Olorunfemi Amoran,
and S.J. Lane. The value of the combined assessment of COPD
in accurate characterization of stable COPD. https://imj.ie/the-value-of-the-combined-assessment-of-copd-in-accurate-characterization-of-stable-copd/,
January 2016.
[STM15] STMicroelectronics. iNEMO inertial module: 3D accelerometer, 3D gyroscope, 3D magnetometer, 3 2015. Rev. 3.
[SWM¸21] Sarah Bettina Schwarz, Wolfram Windisch, Daniel Sebastian Majorski, Jens Callegari, Marilena Pläcking, and Friederike Sophie Magnet. Long-term auscultation in chronic obstructive pulmonary disease: Renaissance of an ideograph of medical care. Respiration, 100(3):201–208, 02 2021.
[Tex18] Texas Instruments. HDC2080 Low-Power Humidity and Temperature Digital Sensor, 7 2018.
[u-b20] u-blox. LARA-R2 series, Size-optimized LTE Cat 1 modules in single and multi-mode configurations, 7 2020.