(Đồ án hcmute) development of a wearable ai based device for wrist pulse diagnosis

ix LIST OF FIGURES Figure 1.1 Wrist pulse diagnosis in traditional medicine Source: https://beta.parkwayshenton.com/healthplus/article/tcm-misconceptions .... Figure 1.1 Wrist pulse dia

MINISTRY OF EDUCATION AND TRAINING

HO CHI MINH CITY UNIVERSITY OF TECHNOLOGY AND EDUCATION

FACULTY OF MECHANICAL ENGINEERING

GRADUATION PROJECT ROBOTIC AND ARTIFICIAL INTELLIGENCE

DEVELOPMENT OF A WEARABLE AI-BASED DEVICE

FOR WRIST PULSE DIAGNOSIS

ADVISOR: PhD BUI HA DUC STUDENT: DAO THANH QUAN

LE QUOC TUAN

S K L 0 1 0 8 3 7 n

Ho Chi Minh city, July 2023

MINISTRY OF EDUCATION AND TRAINING

HO CHI MINH CITY UNIVERSITY OF TECHNOLOGY AND EDUCATION

FACULTY OF MECHANICAL ENGINEERING

GRADUATION THESIS

Year Of Admission: 2019 - 2023

Ho Chi Minh City, July 2023

DEVELOPMENT OF A WEARABLE AI-BASED DEVICE

FOR WRIST PULSE DIAGNOSIS

n

HO CHI MINH CITY UNIVERSITY OF TECHNOLOGY AND EDUCATION

FACULTY OF MECHANICAL ENGINEERING DEPARTMENT OF MECHATRONICS

GRADUATION THESIS

DEVELOPMENT OF A WEARABLE AI-BASED DEVICE

FOR WRIST PULSE DIAGNOSIS

Giảng viên hướng dẫn: TS Bùi Hà Đức……….……… ………

Sinh viên thực hiện: Lê Quốc Tuấn………MSSV: 19134091…Điện thoại 0935920245

Đào Thanh Quân………MSSV: 19134081…Điện thoại 0962600379

1 Mã số đề tài: 22223DT95

– Tên đề tài:

Development of a Wearable AI-based Device for Wrist Pulse Diagnosis

2 Các số liệu, tài liệu ban đầu:

……….……… ……….………

……….……… ……….………

3 Nội dung chính của đồ án:

Xây dựng thiết bị đo và phần mềm xử lý tín hiệu mạch đập ở cổ tay, qua đó chẩn đoán tình trạng sức khỏe của một người

……….……… ……….………

4 Các sản phẩm dự kiến

Thiết bị đo mạch đập cổ tay

Phần mềm phân tích chẩn đoán tình trạng sức khỏe người dựa trên tín hiệu mạch đập cổ tay

……….……… ……….………

……….……… ……….………

5 Ngày giao đồ án: 15/03/2023

6 Ngày nộp đồ án: 15/07/2023

7 Ngôn ngữ trình bày: Bản báo cáo: Tiếng Anh Tiếng Việt

Trình bày bảo vệ: Tiếng Anh Tiếng Việt

(Ký, ghi rõ họ tên) (Ký, ghi rõ họ tên) (Ký, ghi rõ họ tên)

n

COMMITMENT

• Project: Development of a Wearable AI-based Device for Wrist Pulse Diagnosis

• Lecturer: Bui Ha Duc, Ph.D

• Student: Dao Thanh Quan

• Adress: Phuoc Long B Ward, Thu Duc City, Ho Chi Minh City

• Phone number: 0962600379

• Email: quandaoforwork@gmail.com

• Student: Le Quoc Tuan

• Adress: Linh Chieu Ward, Thu Duc City, Ho Chi Minh City

• Phone number: 0935920345

• Email: lequoctuan.fwork@gmail.com

• Graduation thesis submission date:

• Commitment: “I affirm that the graduation thesis presented here is the result of my research and efforts I have not replicated any content from published articles without appropriate citations Should any breach be identified, I acknowledge full accountability for the consequences.”

Ho Chi Minh City, … July 2023 n

iii

ACKNOWLEDGMENT

First and foremost, on behalf of my team, I would like to extend my heartfelt thanks to our supervisor, Bui Ha Duc Ph.D Your unwavering commitment, expertise, and guidance have been instrumental in shaping my research and steering me toward success Your mentorship, patience, and constant encouragement have played a significant role in shaping my academic journey Your profound knowledge and invaluable insights have helped me overcome challenges and broaden

my intellectual horizons I am truly grateful for your dedication and support

I would also like to extend my sincere appreciation to Ho Chi Minh City University of Technology and Education for providing me with an exceptional learning environment and resources The university has been a constant source of inspiration, fostering an atmosphere of growth, innovation, and academic excellence The diverse faculty and staff have been instrumental in shaping my academic and personal development, and I am indebted to them for their commitment to nurturing future scholars and leaders

My deepest gratitude goes to my family for their unwavering love, encouragement, and understanding Your support has been the foundation of my journey, and I am forever grateful for your sacrifices, belief in me, and the countless ways you have cheered me on Your unwavering support has provided me with the strength and motivation to overcome challenges and strive for excellence

Lastly, I would like to express my gratitude to all those who have contributed to my growth and development, directly or indirectly Your encouragement, advice, and belief in my abilities have been invaluable throughout this challenging yet rewarding journey

Sincerely, Dao Thanh Quan n

ABSTRACT

Pulse diagnosis, an integral part of traditional medicine (TM), is a crucial diagnostic method alongside looking, listening, and asking It involves the practitioner placing their three fingers on the patient's radial artery at the wrist to analyze their health condition This method has been used for thousands of years in TM and continues to be highly regarded for its convenience, affordability, and non-invasive nature Pulse diagnosis remains a strong contender for disease diagnosis even in modern times

Recent studies have highlighted the significance of the wrist pulse signal as a bloodstream signal that can provide valuable insights for disease analysis However, traditional pulse diagnosis (TPD) heavily relies on the expertise of practitioners The measurement and interpretation involved in TPD typically demand years of training for practitioners Additionally, there is a lack of standardized communication and sharing of pulse signal experiences among different practitioners These factors present challenges in the further development and widespread application of TPD in modern clinical practice

Advancements in sensor technologies, signal processing, and pattern recognition have paved the way for significant advancements in the computational analysis of pulse signals These developments have led to the creation of three types of sensors for pulse signal acquisition: pressure sensors, photoelectric sensors, and ultrasonic sensors These sensors enable the simulation of pulse signal analysis, resembling the expertise of practitioners Signal processing and pattern recognition methods have been devised to interpret and analyze pulse signals As a result, pulse signals have been extensively investigated for various applications, including pulse waveform classification, prediction, and the diagnosis of numerous diseases such as cholecystitis, nephrotic syndrome, diabetes, and more

The objectives of this study are to research, design, and fabricate a measuring device equipped with a piezo sensor to capture wrist pulse signals, and apply artificial intelligence to analyze the recorded signal In addition, we have gathered data on wrist pulse (using our developed device) and blood glucose levels (from a commercially available device) from a group

of individuals over several days, with measurements taken at various times throughout the day

By applying digital signal processing techniques, we have effectively eliminated noise sources such as high-frequency noise, electrostatic noise, and baseline wander from the collected data Our ultimate goal is to leverage this processed data to create an AI model capable of predicting pulse signals and blood glucose levels Our research findings have been accepted and approved

at the 2023 International Conference on System Science and Engineering (ICSSE)

n

v

TABLE OF CONTENTS

NHIỆM VỤ ĐỒ ÁN TỐT NGHIỆP i

COMMITMENT ii

ACKNOWLEDGMENT iii

ABSTRACT iv

TABLE OF CONTENTS v

LIST OF TABLES viii

LIST OF FIGURES ix

LIST OF ACRONYMS xii

CHAPTER 1.INTRODUCTION 1

1.1 Motivation 1

1.2 Scientific and practical significances 2

1.3 Objectives 2

1.4 Research methods 2

1.5 Structure of the report 3

CHAPTER 2.LITERATURE REVIEW 4

2.1 Introduction to Traditional Medicine 4

2.2 Introduction to the use of sensors in collecting pulse wave signals 7

2.2.1 Photoplethysmography (PPG) sensors 8

2.2.2 Piezoresistive sensors 9

2.2.3 Capacitive pressure sensors 11

2.2.4 Piezoelectricity sensors 13

2.3 Research overview 15

2.4 Introduction to noise-removing technique used for pulse signal 17

2.4.1 Finite Impulse Response (FIR) filter 18

2.4.2 Infinite Impulse Response (IIR) filter 19

2.4.3 Wavelet-based filter 20

2.5 Introduction to metrics used in time series 21

2.5.1 Introduction to signal-to-noise ratio 21

2.5.2 Introduction to mean squared error 22

2.6 Introduction to Machine Learning in Traditional Medicine 22

2.6.1 Application of machine learning 22

2.6.2 Introduction to statistical model used in time series forecasting 23

n

2.6.3 Introduction to deep learning model used in time series forecasting 25

CHAPTER 3.DESIGN MECHANICAL SYSTEM 29

3.1 Hardware objectives 29

3.2 Technical requirements 29

3.3 Design options 30

3.3.1 Design the bracelet 30

3.3.2 Design the electrical box 38

CHAPTER 4.DESIGN ELECTRONICS – CONTROL SYSTEM 40

4.1 Electronics – control system’s objectives 40

4.2 Technical requirements 40

4.3 Design options 41

4.3.1 Cetral control block 41

4.3.2 Power supply block 41

4.3.3 Sensor block 42

4.3.4 Signal processing block 42

4.4 Hardware experiment 48

4.4.1 Experiment set up 48

4.4.2 Experimental result 50

CHAPTER 5.DESIGN PULSE WAVE FORECASTING ALGORITHM 53

5.1 Objectives 53

5.2 Data Preparation 53

5.2.1 Equipment 53

5.2.2 Patient recruitment and protocol 53

5.3 Preprocessing algorithm 56

5.3.1 Applying wavelet transform in filtering signal 56

5.3.2 Baseline wander removal 59

5.4 Time series forecasting model 62

5.4.1 Time2Vec 63

5.4.2 Gating mechanism 63

5.4.3 Feature selection network 66

5.4.4 Interpretable multi-head attention 67

5.4.5 Locality enhancement 68

5.5 Experiment 68

5.5.1 Training procedure 68

n

vii

5.5.1 Experimental result 69

CHAPTER 6.BLOOD GLUCOSE MEASUREMENT 72

6.1 Objectives 72

6.2 Blood glucose measuring model 73

6.3 Experiment 75

6.3.1 Dataset preparation 75

6.2.5 Training procedure: Blood glucose measuring model 76

6.2.6 Experimental result: Blood glucose measuring model 77

CONCLUSION AND DISCUSSION 78

REFERENCE 79 APPENDIX 1 I

n

LIST OF TABLES

Table 3.1: Common 3D printer filaments comparison 35

Table 4.1: High-Precision AD HAT specifications 44

Table 4.2: High-Precision AD HAT Pinouts 45

Table 5.1: The comparison between forecasting models 70

n

ix

LIST OF FIGURES

Figure 1.1 Wrist pulse diagnosis in traditional medicine (Source:

https://beta.parkwayshenton.com/healthplus/article/tcm-misconceptions) 1

Figure 2.1: Traditional Medicine methods 4

Figure 2.2: Tongue describes organ system map on tongue (Source: https://www.wellnessprinciple.com/blog/traditional-chinese-medicine-tongue-diagnosis) 5

Figure 2.3: The positions of Cun – Guan – Chi and their equivalent organs 7

Figure 2.4: PPG sensors placements 8

Figure 2.5: PPG reflection type working principle 8

Figure 2.6: Connected circuit between Piezoresistive sensor and Wheatstone Bridge (Source: https://www.avnet.com/wps/portal/abacus/solutions/technologies/sensors/pressure-sensors/core-technologies/piezoresistive-strain-gauge/) 10

Figure 2.7: Capacitive pressure sensor structure (Source: https://www.avnet.com/wps/portal/abacus/solutions/technologies/sensors/pressure-sensors/core-technologies/capacitive/) 12

Figure 2.8: Piezoelectric pressure sensor construction 14

Figure 2.9: Bi-Sensing Pulse Diagnosis Instrument and holder of pulse-taking posture (Source: [13]) 16 Figure 2.10: (a) Proposed system, (b) Real photograph of the system with real-time UI 16

Figure 2.11: Signal with power line interference 17

Figure 2.12: Signal with baseline wander 18

Figure 2.13: A direct form discrete-time FIR filter of order N (Source: Wikipedia) 19

Figure 2.14: Wavelet Families of discrete wavelets and continuous wavelets (Source: [4]) 21

Figure 2.15: Non-stationary and stationary series example (Source: https://otexts.com/fpp2/arima.html) 24

Figure 2.16: Visualization of LSTM Architecture (Source: https://thorirmar.com/post/insight_into_lstm/) 27

Figure 2.17: Transformer visualization (Source:[2]) 28

Figure 3.1: Wrist pulse wave diagnosis method 29

Figure 3.2: (a) Penetrative method, (b) None-penetrative method (Source: https://www.sparkfun.com/datasheets/Sensors/Flex/MSI-techman.pdf) 30

Figure 3.3: CAM mechanism description and design 31

Figure 3.4: Visualization of the response signals on Labview 31

Figure 3.5: The first design option 32

n

Figure 3.6: The second design option 33

Figure 3.7: The modified isolator element 34

Figure 3.8: (a) The bracelet’s frame, (b) Vertical half-view of the isolator element 34

Figure 3.9: The proposed assembly design of the system 35

Figure 3.10: Printing process on Elegoo Neptune Plus 3 3D Printer Machine 37

Figure 3.11: The completed bracelet 37

Figure 3.12: The proposed design of the electrical box 38

Figure 3.13: The completed electrical box 39

Figure 4.1: Block diagram of electronics – control system 40

Figure 4.2: Power supply adaptor for Raspberry Pi (Source: Internet) 41

Figure 4.3: Piezoelectric sensor (ID: DT1-052K) 42

Figure 4.4: Raspberry Pi Zero (source: internet) 43

Figure 4.5: Onboard High-Precision AD HAT (source: https://www.waveshare.com/18983.htm) 44

Figure 4.6: Serial Interface Timing Requirements (Source: https://www.ti.com/document-viewer/ads1263/datasheet ) 46

Figure 4.7: Serial Interface Switching Characteristics (Source: https://www.ti.com/document-viewer/ads1263/datasheet) 47

Figure 4.8: Data red directly by ADC1(Source: https://www.ti.com/document-viewer/ads1263/datasheet) 48

Figure 4.9: Electrocardiogram sensor SEN0213 (Source: digikey.com) 49

Figure 4.10: Block diagram of an additional circuit for device validation 50

Figure 4.11: 5-second filtered signal of ECG and the proposed device on subject 1 50

Figure 4.12: 5-second filtered signal of ECG and the proposed device on subject 2 51

Figure 4.13: 5-second filtered signal of ECG and the proposed device on subject 3 51

Figure 4.14: 5-second filtered signal of ECG and the proposed device on subject 1 52

Figure 5.1: Important features inside 2 consecutive pulse wave 54

Figure 5.2: The boxplots showing distribution of heart rate variability, pulse period, diastolic peak height, systolic peak height on 4 different times including Before Breakfast (BB), After Breakfast (AB), Before Lunch (BL), After Lunch (AL) 55

Figure 5.3: Data acquisition demonstration 55

Figure 5.4: Frequency bands division of wavelet transform 57

Figure 5.5: ‘Sym5’ wavelet used in the proposed algorithm 58

n

xi

Figure 5.6: The comparisons between raw and filtered data on 5 different subjects with each row

belongs to a subject 58

Figure 5.7: The positions of pulse’s onset and systolic peak inside pulse wave 59

Figure 5.8: Pulse’s onset detection algorithm 60

Figure 5.9: Baseline wander removal algorithm 61

Figure 5.10: Block diagram of the forecasting model 62

Figure 5.11: Block diagram of the gated residual network 65

Figure 5.12: Block diagram of gated linear unit 66

Figure 5.13: Block diagram of feature selection network 67

Figure 5.14: Results of iterative inference on 4 different subjects 71

Figure 6.1: The global number of people with diabetes in 2017 and 2045 (Source: IDF Diabetes Atlas 8th edition) 72

Figure 6.2: Visualization of fluid compartments (Source: https://courses.lumenlearning.com/suny-ap2/chapter/body-fluids-and-fluid-compartments-no-content/) 73

Figure 6.3: Block diagram of blood glucose measuring model 74

Figure 6.4: Zephyr Bioharness 3 device used in D1NAMO project (Source: [2]) 76

n

n

xiii

CHAPTER 1 INTRODUCTION 1.1 Motivation

For thousands of years, pulse diagnosis has played an important role in traditional medicine for disease analysis This is a well-known noninvasive assessment approach that is successfully used in many Eastern countries In this method, doctors place their fingers on the wrist of the patients, near the radius bones They then apply several levels of force to feel the responding force created by the blood vessels There are several properties that the doctors want to feel, such

as pulse rate, strength, width, etc., created by the blood vessels in the wrist In general, the doctors use these properties to assess the overall balance inside the patient and come up with their diagnosis

In the past, it took traditional medicine practitioners many years of hard work and practice

to accumulate their experience However, there is hardly any formal document teaching this ancient technique, so their knowledge might not be standardized Besides, the diagnostic results subjectively depend on the experience of the doctors, so they may not be consistent between doctors Additionally, the feelings of a doctor can be different from time to time, so the results may also be varied even if they are made by the same doctor Moreover, by relying on their experience, it might be hard for the practitioners to explain the result to their patients

Besides, a pulse diagnosis process happens in a short period of time This may lead to many disadvantages, such as the practitioners losing their concentration and missing important information at certain points This can significantly affect the diagnostic result Because the pulse wave data cannot be saved, the practitioners cannot do any deeper analysis to deliver better results and cannot create medical records for the patients

Figure 1.1 Wrist pulse diagnosis in traditional medicine (Source:

https://beta.parkwayshenton.com/healthplus/article/tcm-misconceptions)

n

2

Additionally, telemedicine is getting more and more attractive around the world because of the development of technology It can not only save time for the doctors and patients but also make it accessible to many people, especially those from remote areas Telemedicine can also reduce hospital fees while still enhancing the quality of medical services However, in the modern approach, clinical examinations require many modern machines, that are extremely large and expensive, in order to deliver an accurate result It becomes a barrier that prevents telemedicine from conducting and developing Hence, with the traditional medicine approach, telemedicine can overcome these limits and thrive

Realizing the above challenges and opportunities, we decide to research and develop a wearable device that can record the pulse wave from the wrist, based on the principles of traditional medicine Moreover, the device is facilitated with modern technologies to convert the recorded signal into valuable information used to monitor humans health

1.2 Scientific and practical significances

• Scientific significances: The study establishes the researching and developing artificial intelligence applications in the healthcare field, with the aim of improving quality of life The self-collected dataset can serve as a valuable resource for further studies on AI applications in diagnosis and alerting of potential risks in diabetes

• Diagnosing and maintaining patient medical records

1.3 Objectives

We have a strong desire to contribute to comprehensive research in Traditional Medicine

as a whole, with a specific focus on its application in the management of diabetes

• Develop a compact and user-friendly device capable of accurately capturing pulse wave signals

• Create a dataset consisting of pulse wave signals and blood glucose levels (BGL) to serve the project’s objectives and make it publicly available for future research

• Forecast future pulse signals to estimate future blood glucose levels and provide alerts for abnormal symptoms Extract and visualize the factors influencing the prediction results

• Conduct a comprehensive study on digital signal processing techniques and apply appropriate methods to the wrist pulse signals in order to obtain a complete dataset

• Study, analyze, and evaluate deep learning models to select a model that aligns with the project's requirements Apply experimental methods to conduct real-world testing, draw conclusions, and progressively implement and refine the project based on the researched and summarized theoretical foundation

1.5 Structure of the report

The report includes 6 chapters:

Chapter 1 INTRODUCTION: Brief introduction to the study

Chapter 2 LITERATURE REVIEW: Theoretical knowledge related to the study: Traditional

Medicine, sensors, pulse wave analysis, deep learning models in Traditional Medicine

Chapter 3 DESIGN MECHANICAL SYSTEM: Design and fabricate the mechanical system Chapter 4 DESIGN ELECTRONICS-CONTROL SYSEM: Explain the block diagram in the

electronics-control system

Chapter 5 DESIGN AI ALGORITHM: Objectives of the AI algorithm, step by step preprocess

data and build the AI models

And the last chapter is the conclusion and discussionn

4

2.1 Introduction to Traditional Medicine

In Traditional Medicine, there are four diagnostic methods used to examine human health conditions namely inspection, auscultation, olfaction and interrogation, and palpation Each method is based on different information such as external conditions, family history, diets, living

states, etc of the patients to examine their wellness [7]

Figure 2.1: Traditional Medicine methods

Firstly, the inspection involves observing and examining the patient's physical appearance, including their complexion, facial expressions, body movements, and posture In TM, the practitioner pays attention to the color and shape of the tongue, the condition of the skin, and the presence of any abnormal physical characteristics These observations provide insights into the patient's overall health, organ function, and potential imbalances For example, in terms of tongue status, it is believed that the tongue reflects the reserve map of the body’s torso The tongue is divided into different sections that correspond to various organ systems and emotions in traditional medicine By examining the appearance of each section, valuable information about the health of these organs can be obtained The tip of the tongue is associated with the heart, lungs, and emotions Observing the tip can provide insights into the condition of these organs and reveal emotional imbalances The center of the tongue is connected to the digestive system

By examining this area, practitioners can gather information about the health of the stomach, spleen, and intestines, aiding in the diagnosis of digestive disorders The sides of the tongue are believed to reflect the condition of the liver and gallbladder Changes or abnormalities in these

n

areas may indicate issues with liver function or the presence of gallbladder-related problems Lastly, the base of the tongue is associated with the kidneys, bladder, and intestines Observing this region can offer clues about the state of these organs and help identify any underlying imbalances or diseases

Figure 2.2: Tongue describes organ system map on tongue (Source:

https://www.wellnessprinciple.com/blog/traditional-chinese-medicine-tongue-diagnosis)

Secondly, auscultation refers to the act of listening to sounds produced by the body, which can provide valuable insights into the internal state of organs and help in diagnosing specific diseases or imbalances During auscultation in TM, the practitioner listens to various sounds produced by the body, such as the quality of a patient's breath, voice, or the sounds emanating from the abdomen These sounds are considered indicators of the functioning of internal organs and the overall balance of the body's systems For example, in TM, the sound of the breath can

be observed to determine the quality and depth of respiration Abnormal breath sounds, such as wheezing or rattling, may indicate imbalances or pathologies in the lungs or respiratory system Similarly, the practitioner may listen to the patient's voice, as changes in the tone, volume, or clarity of speech can provide insights into the condition of the throat, vocal cords, or other related organs Additionally, the sounds produced by the abdomen during digestion, such as rumbling

or gurgling, can be assessed to gain information about the state of the digestive system, including

n

6

the stomach and intestines Auscultation in TM is performed by trained practitioners who have developed the ability to discern subtle variations in sounds and interpret them within the framework of TM theory It is often used in conjunction with other diagnostic methods, such as inspection, inquiry, and palpation, to form a comprehensive understanding of the patient's health condition and guide treatment strategies

Olfaction refers to the act of smelling odors emitted by the patient, which can provide valuable insights into the internal state of the body and aid in diagnosing specific diseases or imbalances During olfaction in TM, the practitioner pays attention to specific odors that may be present on the patient's body, breath, or excretions These odors are believed to be indicative of underlying imbalances or pathologies in the body For example, certain breath odors can be associated with specific organ dysfunctions A foul breath odor, such as a "rotten egg" smell, may indicate issues with the digestive system, while a sweet or fruity breath odor could be linked

to imbalances in blood sugar levels Olfaction can also be used to assess body odors, which may reveal imbalances related to sweat glands, skin conditions, or metabolic processes Specific odors emanating from the skin, urine, or stools can provide clues about the functioning of various organs or systems within the body In TM, the sense of smell is considered an important diagnostic tool, and practitioners trained in olfaction can discern subtle variations in odors and interpret them within the framework of TM theory Whereas, interrogation, also known as inquiry

or questioning, involves a series of detailed questions asked by the practitioner to the patient During the interrogation process in TM, the practitioner aims to gather comprehensive information about the patient's symptoms, medical history, lifestyle, diet, emotional well-being, and other relevant factors This line of questioning helps the practitioner understand the patient's overall health and identify patterns or imbalances that may be contributing to their condition

In this study, we mainly focus on wrist pulse wave diagnosis, which belongs to palpation methods The wrist pulse diagnosis is an important method to assess a person’s overall health conditions and identify imbalances within the human body It is commonly believed by TM practitioners that wrist pulses can reflect the flow of qi, which is a vital energy or life force that everything in the world is made up of qi Qi circulates throughout the human body, as well as every organ inside it, through meridian channels This flow is then reflected in many acupuncture points in the body, including cun, guan, and chi positions Hence, wrist pulse diagnosis is a way

to capture information about qi through acupuncture points According to TM theory, the stability and balance of the flow of qi are the most important concepts used to judge the wellness of the human body By assessing the flow of qi through the wrist pulse, TM practitioners can grasp the

whole information about the internal organs of a person [8] Moreover, many recent studies have

n

given scientific proof of this kind of diagnosis In detail, during the respiratory process, the heart creates an oscillation that is distributed throughout the arterial system of the human body Side-branch organs will then react to the oscillation, generate harmonic forces with maximum amplitudes near their own natural frequencies, and then distribute them back to the arterial system Based on this “frequency match” theory, it is recorded that the frequencies at several

acupuncture points match the natural frequencies of several organs [4, 6]

To assess the wrist pulse, TM practitioners simultaneously place their index, middle, and ring fingers on the radial artery of the patient, which correspond to cun, guan, and chi, respectively These positions are commonly believed to represent the states of internal organs of the human body For example, the cun, guan, and chi positions on the left hand correspond to heart, liver, kidney while those on the right hand correspond to lung, stomach and kidney, respectively Because the pulse varies according to applied static forces, practitioners will then respectively apply light, medium, heavy forces to feel for various qualities, including pulse rate, trend, strength, length, width, etc Based on these qualities, practitioners with experience can tell

the imbalance and overall health state of a person [1, 11, 13].

Figure 2.3: The positions of Cun – Guan – Chi and their equivalent organs

2.2 Introduction to the use of sensors in collecting pulse wave signals

The use of sensors in collecting pulse wave signals has become increasingly prevalent in modern healthcare Pulse wave sensors are typically non-invasive devices that are designed to measure and detect the change in arterial pulse These sensors are integrated into various

n

8

wearable devices or medical equipments, allowing for continuous monitoring of pulse wave signals There are different types of sensors used to collect pulse wave signals, including photoplethysmography (PPG) sensors, laser doppler sensors, piezoresistive sensors, piezoelectric sensors, etc

2.2.1 Photoplethysmography (PPG) sensors

PPG sensors are non-invasive devices that utilize light-emitting diodes (LED) and photodetectors to measure variations in blood volume and flow They are often placed on the skin, typically at the fingertip or wrist, to capture pulse wave signals

PPG sensors work based on the principle of optical absorption and reflection The sensor emits light from an LED onto the skin, and the photodetector measures the intensity of the light that is either absorbed or reflected by the underlying blood vessels This variation in light intensity is caused by changes in blood volume with each heartbeat, resulting in a pulsatile waveform known as the PPG signal

Figure 2.5: PPG reflection type working principle

Figure 2.4: PPG sensors placements

n

• Cost-Effective: PPG sensors are generally more cost-effective compared to invasive methods or more complex diagnostic equipment

• Portable and Wearable: PPG sensors can be integrated into wearable devices or mobile applications, enabling convenient and portable monitoring for personal health tracking

• Subject to Environmental Factors: External factors such as ambient light, temperature, and skin pigmentation can influence the performance and accuracy of PPG sensors

2.2.2 Piezoresistive sensors

Piezoresistive strain gauges are widely used as pressure sensors due to their commonality and effectiveness These sensors operate based on the principle that the electrical resistance of a material changes when it undergoes deformation or strain The fundamental concept behind a piezoresistive pressure sensor involves utilizing a strain gauge made of a conductive material that exhibits changes in electrical resistance when subjected to stretching This strain gauge is typically affixed to a diaphragm, which acts as the sensing element and deforms in response to applied pressure The sensitivity of each material is calculated by a gauge factor, which is defined

as the ratio of the relative resistance change and the strain

𝐶𝐹 =

(𝛥𝑅𝑅 )𝜖

(2.1) Where strain is defined as the relative change in length:

n

10

𝜖 =𝛥𝐿

The strain gauge elements can be made of various materials, including metals such

as constantan (a copper-nickel alloy), karma alloy (nickel-chromium), nickel, and semiconductors such as silicon, germanium, or polysilicon While metal materials offer good mechanical stability, semiconductor strain gauges often have a higher gauge factor than metal foils, providing enhanced sensitivity However, they can be more sensitive to temperature variations and require careful temperature compensation In real-life applications, a Wheatstone bridge circuit is commonly employed to measure the change in resistance in the strain gauge sensor This circuit configuration enables the conversion of small variations in the sensor's resistance into an output voltage

Figure 2.6: Connected circuit between Piezoresistive sensor and Wheatstone Bridge (Source: https://www.avnet.com/wps/portal/abacus/solutions/technologies/sensors/pressure-sensors/core-

technologies/piezoresistive-strain-gauge/)

The Wheatstone bridge circuit requires an excitation voltage to be applied In the absence

of strain, when all resistors in the bridge are balanced, the output voltage will be zero However, when pressure is applied, causing resistance changes in the bridge, an output voltage or current will be generated The calculation for determining this output is provided by the formula below

• High Sensitivity: Piezoresistive sensors exhibit high sensitivity, allowing them to detect even small changes in pressure or strain

• Fast Response Time: Piezoresistive sensors have a fast response time, enabling real-time monitoring and measurement of dynamic pressure changes

• Cost-Effective: Piezoresistive sensors are relatively cost-effective compared to other pressure sensing technologies, making them accessible for a wide range of applications and industries

• Compatibility with Electronics: Piezoresistive sensors can be easily integrated with electronic circuits and microcontrollers for signal processing and data acquisition They can be combined with digital interfaces, allowing for easy integration into larger systems

- Disadvantages of Piezoresistive sensors:

• Temperature Sensitivity: Piezoresistive sensors can be sensitive to temperature changes, which may affect their accuracy and require temperature compensation techniques to ensure reliable measurements

• Non-Linearity: The output response of piezoresistive sensors may exhibit linear behavior, especially at higher pressure ranges Calibration may be necessary

non-to obtain accurate and linear measurements

• Drift: Over time, piezoresistive sensors can experience drift in their output readings, requiring periodic recalibration to maintain accuracy

• Mechanical Fragility: Piezoresistive sensors can be sensitive to mechanical stress and may be prone to damage if exposed to excessive force or shocks Proper handling and protection are necessary to ensure their longevity

2.2.3 Capacitive pressure sensors

Capacitive pressure sensors are devices that utilize changes in capacitance to measure and monitor pressure variations These sensors consist of two conductive plates or electrodes separated by a dielectric material The capacitance between the plates is directly influenced by the applied pressure, causing a change in the capacitance value The capacitance is defined by:

12

• 𝜀0 is the electric constant (equal to 8.854x1012 F/m),

• A is the area of the plates

• d is the distance between the plates

Capacitive pressure sensors typically consist of a diaphragm or membrane that deforms in response to applied pressure The diaphragm acts as one of the electrodes, while the other electrode is a fixed plate or backplate The space between the electrodes is filled with a dielectric material When pressure is applied to the diaphragm, it deforms, altering the distance between the two electrodes This change in spacing results in a change in capacitance, as the effective area and separation between the electrodes are modified The capacitance increases with decreased spacing and vice versa (following to the above equation) The variation in capacitance is measured using appropriate electronics and signal processing techniques The sensor is connected to an external circuit, which applies an alternating current (AC) or direct current (DC) voltage to the sensor The resulting capacitance change is converted into an output signal, typically voltage or frequency, that represents the applied pressure

Figure 2.7: Capacitive pressure sensor structure (Source:

https://www.avnet.com/wps/portal/abacus/solutions/technologies/sensors/pressure-sensors/core-technologies/capacitive/)

- Advantages of capacity pressure sensors:

• High Sensitivity: Capacitive sensors can achieve high sensitivity, allowing them to detect small pressure changes accurately

• Wide Measurement Range: Capacitive pressure sensors can be designed to measure a wide range of pressures, from very low to high values

n

• Good Linearity: These sensors often exhibit good linearity, meaning the output response is proportional to the applied pressure, allowing for precise measurements

• Low Power Consumption: Capacitive pressure sensors typically consume low power, making them suitable for battery-operated devices and applications where power efficiency is essential

- Disadvantages of capacity pressure sensors:

• Temperature Sensitivity: The capacitance of the sensor can be influenced by temperature variations, necessitating temperature compensation techniques for accurate measurements

• Limited Overload Capacity: Exceeding the maximum pressure limit can damage the diaphragm or affect the sensor's performance

• Susceptibility to Environmental Factors: External factors like humidity, moisture, and contaminants can affect the dielectric properties and introduce measurement errors

2.2.4 Piezoelectricity sensors

Piezoelectric sensors are electronic devices that can transform mechanical or thermal energy into electrical signals, utilizing the principle of electromechanical coupling This phenomenon of piezoelectricity occurs in certain materials that generate an electrical voltage when exposed to mechanical stress Conversely, these materials also produce mechanical stress when subjected to an electrical voltage The unique capabilities of piezoelectric sensors make them particularly well-suited for applications in Traditional Medicine (TM)

One significant advantage of piezoelectric sensors is their exceptional sensitivity They can detect even the smallest pressure waves, such as the pulse wave generated by the heartbeat This high sensitivity allows for precise monitoring and analysis of physiological signals related to the pulse By accurately capturing and interpreting these subtle pressure variations, piezoelectric sensors contribute valuable insights into a person's health and well-being

In addition to their sensitivity, piezoelectric sensors boast an impressive response time The pulse wave is a dynamic signal that changes rapidly, requiring sensors with a fast response time

to capture these fluctuations accurately Piezoelectric sensors excel in this aspect, providing time data that is crucial for assessing pulse characteristics in TM The ability to capture rapid changes in pressure accurately enables practitioners to make informed diagnoses and tailor treatments accordingly

real-n

14

Figure 2.8: Piezoelectric pressure sensor construction

Piezoelectric sensor is constructed with a piezo film which is usually made from Zinc oxide

or Lead zirconate titanate (PZT) These materials are used because of their larger piezoelectric effect When pressure or acceleration is exerted on the PZT material, it generates an equivalent amount of electrical charge across its crystal faces The magnitude of the electrical charge is directly proportional to the applied pressure Generated electrical charge transmitted through metalized plate can be read by an electrical device The protective coating is used to cover the metallization and piezoelectric film inside

- Advantages of Piezoelectric Sensors:

• High Sensitivity: Piezoelectric sensors exhibit high sensitivity, allowing them to detect even small changes in pressure, strain, or acceleration They can provide precise and accurate measurements

• Wide Frequency Range: Piezoelectric sensors are capable of measuring dynamic events with high frequency response They can capture fast-changing signals and vibrations accurately

• Broad Measurement Range: Piezoelectric sensors can be designed to measure a wide range of pressures, forces, and accelerations, making them versatile for different applications

• Rugged and Durable: Piezoelectric sensors are known for their robustness and durability They can withstand harsh environments, high temperatures, and mechanical stress without significant loss in performance

• Fast Response Time: Piezoelectric sensors have a rapid response time, enabling real-time monitoring and measurement of dynamic events

n

• Wide Application Range: Piezoelectric sensors find applications in various fields, including automotive, aerospace, robotics, medical devices, structural analysis, and industrial monitoring

- Disadvantages of Piezoelectric Sensors:

• Limited Linearity: The output response of piezoelectric sensors may exhibit linear behavior, especially at higher input levels Calibration or compensation techniques may be necessary to improve linearity

non-• Temperature Sensitivity: Piezoelectric sensors can be sensitive to temperature variations, which may affect their accuracy Temperature compensation techniques may be required for precise measurements

• Fragile and Delicate: Piezoelectric sensors are relatively fragile and can be prone

to damage if subjected to excessive mechanical stress or shock Careful handling

is necessary to maintain their integrity

• External Power Source: Piezoelectric sensors require an external power source, such as a charge amplifier, to convert their high-impedance output into a usable signal This adds complexity to the measurement setup

• Limited Static Measurements: Piezoelectric sensors are primarily designed for dynamic measurements and may not provide accurate static measurements due to drift and hysteresis effects

• Limited Bandwidth: The frequency response of piezoelectric sensors may be limited at extremely high frequencies, depending on the specific sensor design and materials used

2.3 Research overview

Despite the widespread popularity of Traditional Medicine in Vietnam and its trusted status

as a highly accurate method of medical examination and treatment, there is a lack of research on the application of technology to support hospitals and medical practitioners in utilizing TM methods for diagnosis The current practices in TM hospitals still primarily rely on traditional processes and heavily depend on the expertise and experience of doctors, while the management and recording of medical records remain limited

Furthermore, recognizing the rapid advancement in the application of science and technology in the healthcare field, numerous studies have been published concerning the analysis

of wrist pulse signals For example, Chung et al [13] use three sensors to get the wrist’s pulse

information simultaneously and used Three-Dimension Pulse Mapping to simulate the actual

n

16

touch sensation The device can capture and acquire the finger-reading skills of physicians It is capable of extracting Three Positions Nine Indicators pulse data specifically from Fu-Zhong-Chen displacements located on Cun-Guan-Chi positions

Figure 2.9: Bi-Sensing Pulse Diagnosis Instrument and holder of pulse-taking posture (Source: [13])

Chuangly Chen et al [1] use a sensor array composed of 3 rows x 4 columns MEMS sensors

to record wrist pulse pressure and mapped it to 3D data By obtaining 3D pulse wave data, it is certain that they can extract much more important features from the pulse such as depth, and

Figure 2.10: (a) Proposed system, (b) Real photograph of the system with real-time UI

n

length… These devices open up the potential in the collection of medical data, thereby applying deep learning models to the development of related applications

2.4 Introduction to noise-removing technique used for pulse signal

In pulse wave signal, which is a reflection of the activity of the heart, there are many types

of noises that can affect the quality and accuracy of the recorded response Both physiological or non-physiological factors can be the reason for these types of noises Some common types of noises include:

• Baseline wander: Baseline wander or baseline drift is a low-frequency variation of the signal that can create signal distortion, decrease signal quality or missing detections, etc This type of noise arises from breathing, body movement or poor electrodes connection

• Power line interference: Power line interference is a common type of noise in electrical devices that is created by the power supply to the device or the neighboring power lines or electrical devices It introduces a 50-60 Hz sinusoidal waveform, based on the region, that significantly contaminates the recorded data This contamination can obscure and distort the recorded signal that makes it difficult to accurately analyze and interpret the physiological data

Figure 2.11: Signal with power line interference

n

18

Figure 2.12: Signal with baseline wander

In order to remove these types of noises, many different approaches were proposed such as the finite impulse response filter, infinite impulse response filter, and the wavelet-based filter

2.4.1 Finite Impulse Response (FIR) filter

A FIR is a filter whose impulse response is of finite duration since it reaches zero in a finite amount of time In other words, it is a non-recursive filter that has no feedback in its equation

By not utilizing feedback, FIR filters are inherently stable This stability ensures that the output

of the filter doesn't exhibit any unbounded or oscillatory behavior Additionally, the phase issue

of FIR filter can easily by designed to be linear phase by making the coefficient sequence symmetric; linear phase, or phase change proportional to frequency, corresponds to equal delay

o 𝑥[𝑛] is the input signal,

o 𝑦[𝑛] is the output signal

o 𝑁 is the filter order

o 𝑏𝑖 is the value of the impulse response at the ith instant for 0 ≤ 𝑖 ≤ 𝑁 of an 𝑁𝑡ℎ-order FIR filter

n

Figure 2.13: A direct form discrete-time FIR filter of order N (Source: Wikipedia)

2.4.2 Infinite Impulse Response (IIR) filter

A IIR filter is a recursive filter whose output is computed by using the current and previous inputs and previous outputs This mechanism is called a feedback loop By utilizing it, the IIR filter can achieve more efficient frequency response shaping The transfer function of an IIR filter, which is a polynomial ratio in the complex variable "z" determines the frequency response

of the filter The poles and zeros of the transfer function are the locations on the z-plane where the numerator and denominator polynomials, respectively, equal zero The locations of poles and zeros in the z-plane significantly affect the characteristics of the IIR filter While the former determines the frequency response shape, the latter can cancel or emphasize certain frequencies Based on this, IIR filters can be modified to achieve various frequency response characteristics such as low-pass, high-pass, band-pass, band-stop, by carefully choosing the pole and zero.Mathematically, IIR filter is described as:

𝑦[𝑛] = 1

𝑎0(

𝑏0𝑥[𝑛] + 𝑏1𝑥[𝑛 − 1] + ⋯ + 𝑏𝑃𝑥[𝑛 − 𝑃]

−𝑎1𝑦[𝑛 − 1] − 𝑎2𝑦[𝑛 − 2] − ⋯ − 𝑎𝑄𝑦[𝑛 − 𝑄]) (2.6) Where:

o P is the feedforward filter order

o 𝑏𝑖 are the feedforward filter coefficients

o 𝑄 is the feedback filter order

o 𝑎𝑖 are the feedback filter coefficients

o 𝑥[𝑛] is the input signal

o 𝑦[𝑛] is the output signal

n

By changing translation and dilation constant, wavelet transform becomes an ideal tool that

is used to analyze signals in different frequencies with varying resolutions This process is called multiresolution analysis

The most fundamental and often used wavelet transform is the DWT, which is implemented

by a bank of two-channel filters with varying levels DWT is also known as dyadic wavelet transform since it is created by discretizing the scale and displacement of continuous wavelet transform according to powers of two DWT decomposes the original signal into two parts including approximate coefficients, detail coefficients While approximate coefficients act as a low-pass filter, detail coefficients represent the high-frequency information DWT can be

repeatedly operated on the approximate coefficients to achieve lower resolution components [5]

After each level of decomposition, the frequency bands of the signal are divided by two compared

to the sample rate of the original signal Additionally, the signal is also down-sampled by a factor

of 2 after each level of decomposition

Based on the problem, different types of mother wavelet can be used Some wavelet

families that are common in biomedical signal such as EEG, ECG, or pulse wave include

daubechies, symlet, coiflet, morlet wavelet [3, 15, 16].

n

Figure 2.14: Wavelet Families of discrete wavelets and continuous wavelets (Source: [4])

Normally, researchers have to conduct a comparison based on several metrics such as signal-to-noise ratio or mean squared error to choose the best wavelet families for the problem

2.5 Introduction to metrics used in time series

Whenever we come up with any solution, we have to measure its performance

quantitatively based on a suitable metric In time series, there are many types of metrics used for different purposes such as signal-to-noise ratio, mean squared error

2.5.1 Introduction to signal-to-noise ratio

Signal-to-noise ratio (SNR) is a fundamental concept used in digital signal process, computer vision, artificial intelligence, etc., which is used to measure the ratio of the desired signal power and unexpected signal power It is an important parameter to evaluate the performance and quality of a system that processes, transmits, or filters the signal Although it doesn’t provide detailed information about the specific characteristics of a filters, it provides a measure of the separation of signal and noise based on their relative power levels To calculate SNR, we use the following formulas:

𝑆𝑁𝑅𝑑𝑏= 2𝑙𝑜𝑔10(𝑉𝑆𝑖𝑔𝑛𝑎𝑙

Where:

• 𝑆𝑁𝑅𝑑𝑏: is signal-to-noise ratio calculated in decibel

• 𝑉𝑆𝑖𝑔𝑛𝑎𝑙: is the measured signal voltage

• 𝑉𝑁𝑜𝑖𝑠𝑒: is the measured noise voltage

n

22

2.5.2 Introduction to mean squared error

Mean squared error is a commonly used metric in regression problem It measures the average squared difference between a regression line and its ground truth The higher the mean squared error is, the more difference separated two lines are It is also a popular loss function in regression tasks, since it penalizes larger predictions more heavily due it its squaring operation

To calculate MSE, we use the following formulas:

𝑀𝑆𝐸 =1

𝑛∑(𝑌𝑖− 𝑌̂𝑖)2𝑛

𝑖=1

(2.11)

Where:

o MSE: is mean squared error

o n: is number of data points

o 𝑌𝑖: are observed values

o 𝑌̂𝑖: are predicted values

2.6 Introduction to Machine Learning in Traditional Medicine

2.6.1 Application of machine learning

Time series datasets such as electronic health records (EHR), electrocardiogram (ECG), electroencephalogram (EEG) represent valuable sources of information As for EHR, its information spans a patient’s entire lifetime of care Whether on purpose or not, the document the emergence of new morbidities and comorbidities, the time and stage of diagnosis, the formulation of treatment regimens, and their efficacy They may also capture genetic and lifestyle risk As for ECG and EEG, it represents even more specific information by recording the electrical signal generated by heart and brain activity, respectively By applying machine learning models to the aforementioned datasets, we can achieve a much deeper and more interconnected understanding of individual trajectories of health and disease We can fully utilize the capabilities

of machine learning to develop long-term comprehensive patient management programs that evolve with each individual’s changing context and history and take into account not just one risk but multiple risks once we have a fully quantitative and scientific understanding of the progression of multiple diseases over time Nowadays, machine learning is increasingly attractive in healthcare due to its wide applications Common applications include:

- Dynamic forecasting: Cardiovascular disease, cancer, diabetes are serious and chronic diseases that gradually progress throughout the lifetime of a patient This progress can be

n

segmented into several stages that manifest through clinical observations Dynamic forecasting aims for building a disease progression models from EHR and other informative datasets that can issue personalized dynamic forecasts

- Survival analysis: Survival analysis describes the investigation of the interval between two or more events In particular, it is frequently used to identify risk variables that affect survival, compare the risks of many patients at a certain period of interest, and make judgments regarding the most cost-effective way to gather information

- Screening and monitoring: In clinical examination, the cost for screen and monitoring is expensive However, it is difficult to determine what kind of clinical information to acquire, when to acquire it, and how often to do it for every individual separately Hence, applying machine learning can help optimally balance the cost of acquiring information and the value of the information gained by it

- Early diagnosis: Disease such as cancer, or cardiovascular disease is extremely serious and fatal Hence, the ability to identify them early can lead to better and more effective treatment, and potentially save lives Because we can only have access to the current disease state of a patient, and their disease trajectory is obscure, it is difficult to deliver an accurate diagnosis However, by building a machine learning model to gain a thorough understanding of disease trajectories, we can fully understand information about disease trajectory such as how many stages of disease exist per disease, how individuals may progress through disease stages differently, etc Based on these understandings, we can successfully predict and diagnose disease early on the basis of changing patient characteristics, symptoms, and morbidities

2.6.2 Introduction to statistical model used in time series forecasting

ARIMA, which stands for Autoregressive Integrated Moving Average, is a popular time series forecasting model It is a statistical method used to analyze and predict future values in a time series dataset The ARIMA model combines the autoregressive (AR), integrated (I), and moving average (MA) components to capture the underlying patterns and characteristics of the data

ARIMA model is a combination of 3 components:

- Autoregressive: AR(p) is a regression model with lagged values of y, until p-th time in the past, as predictors The AR component helps capture the temporal dependencies and trends in the data

𝑦̂𝑡 = 𝑐 + 𝜙1𝑦𝑡− 1+ 𝜃2𝑦𝑡− 2+ ⋯ + 𝜃𝑝𝑦𝑡−𝑝+ 𝜀𝑡 (2.12)

n

24

Here, p = the number of lagged observations in the model, ε is white noise at time t, c is a constant and φs are parameters

- Integrated I(d): The difference is taken d times until the original series becomes stationary

A stationary time series is one whose properties do not depend on the time at which the series is observed

Where B is called a backshift operator

Thus, a first order difference is written as:

non-of observation In this example, the order non-of differencing would be one, as the first-order differenced series is stationary

Figure 2.15: Non-stationary and stationary series example (Source: