I need someone to check and rewrite the report as it required in the comments

the whole report is written but there is some comments on the report that needed to be fixed as it mentioned in the photo below

BACHELOR OF SCIENCE DEGREE IN ELECTRICAL ENGINEERING

[Company]

Design and Implementation of a Wearable Carotid Neckband Doppler System for Early Detection of Carotid Artery Stenosis

Report by

Supervisor

[Manager]

Date

[Publish Date]

2

>

DECLARATION STATEMENT

Student Name

Student ID

Signature

Date

1

2

3

ABSTRACT

Carotid artery stenosis can be fatal if not detected in early stage. Carotid artery stenosis occurs when the carotid artery becomes restricted owing to plaque build-up (also known as atherosclerosis). If left untreated, carotid artery stenosis can lead to a variety of complications, including strokes. Diabetic, hypertensive, and high-cholesterol patients are the most vulnerable to this illness. The goal of this research is to create and test a wearable carotid artery neckband that can detect carotid artery stenosis in its early phases. The doppler probe is divided into two halves. The first component is the probe’s head, which will be used for the neckband, and the second part is the probe’s tail, which will be utilized for the waistband. A classification model will be developed based on the collected data to classify or distinguish between signals from healthy subjects and those from patients. A publicly available dataset was cleaned using a custom-made Matlab based application. Bad quality signals will be improved using Cycle Generative Adversarial Network (CycleGan) in SDP II. A 3D model of the probe is designed and simulated. Hardware circuit of the neckband will be implemented in SDP II. A deep learning model will be developed and deployed in server to classify the neckband signal to stratify the patients into healthy and unhealthy.

Design and Implementation of a Wearable Carotid Neckband Doppler System for Early Detection of Carotid Artery Stenosis

\ viii

ACKNOWLEDGEMENTS

TABLE OF CONTENTS

DECLARATION STATEMENT i
ABSTRACT ii
ACKNOWLEDGEMENTS iii
LIST OF FIGURES vi
LIST OF TABLES vii
GLOSSARY viii
Chapter 1 Introduction 1
1.1 Background 1
1.2 Problem Definition 2
1.3 Aims and Objectives 3
1.

4

Design Constraints and Standards 3
1.5 Project Plan 5
1.6 Report Organization 6
Chapter 2 Literature Review 7
2.1 Carotid Artery Stenosis 7
2.2 Detection Techniques 7
2.2.1 X-Ray Imaging 8
2.2.2 Digital Subtraction Angiography (DSA) 8
2.2.3 Magnetic Resonance Imaging (MRI) 8
2.2.4 Duplex Ultrasound 8
2.2.5 Vascular Doppler

9

2.3 Transcranial Doppler Ultrasound

10

2.4 Machine Learning

11

2.5 Stenosis Detection 11
2.5.1 Medical Encoder-Decoder Applications 12
2.5.2 Carotid IMT Segmentation Applications 12
Chapter 3 System Design

14

3.1 Full System Block Diagram 14
3.2 Neckband 3D model

15

3.3 Neckband and Waistband Electric design 15
3.4 Machine Learning Model 15
3.4.1 Dataset Description

16

3.4.2 Signal Extraction 16
3.4.3 Manual Data Cleaning

17

3.5 Performance Metrics

18

Chapter 4 Design Testing and Validation 20
4.1 Doppler Simulation Results 20
4.1.1 Crystal Oscillator Transmitter 20
4.1.2 Signal Receiver

21

4.1.3 Filter Circuit

22

4.1.4 Audio Amplifier

23

4.1.5 Power Supply Circuit

24

4.2 Compliance with Design Constraints

25

Chapter 5 Conclusion & Future Work 26
5.1 Conclusion 26
5.2 Recommendations for Future Work 26
REFERENCES 27

LIST OF FIGURES

Figure 1: Anatomy of Carotid Artery

‎

[3]. 1

Figure 2

: Machine Learning vs Deep Learning

‎

[5].

2

Figure 3

: Carotid Artery Stenosis

‎

[9].

3

Figure 4

: Duplex Ultrasound

‎

[21].

9

Figure 5

: Ultrasonic transducer (probe), PVC container, connector cable,

and printed circuit board for a detector

‎

[22].

10

Figure 6

: Doppler signal collection from middle cerebral artery (MCA) and Internal Carotid Artery (ICA).

11

Figure 7

: Full system block diagram.

14

Figure 8

: Neckband 3D model.

15

Figure 9

: Doppler Probe.

15

Figure 10

: Overview of the method and materials used in this study to classify “Healthy” and “ICU” patients.

16

Figure 11

: Graphical presentation of high quality, low quality, mediocre and corrupted signals.

18

Figure 12

: Dataset details before and after manually labeling and cleaning.

18

Figure 13:Simulation of Crystal Oscillator Transmitter Circuit using Multisim. 20

Figure 14

: The output waveform of the transmitter circuit.

21

Figure 15

: Balanced Modulator & Demodulator circuit connection.

21

Figure 16

: Balanced Modulator and Demodulator circuit output.

22

Figure 17

: Simulation of Bandpass Filter circuit.

22

Figure 18

: Bode plots of a Bandpass filter.

23

Figure 19

: Audio Amplifier multisim circuit.

23

Figure 20

: Output of audio amplifier circuit.

23

Figure 21

: Power supply circuit simulation.

24

LIST OF TABLES

Table

‎
1
1

: List of the system constraints.

4

Table
‎
1
2

: List of the system standards.

4

Table
‎
3
1

: Algorithm Envelope tracing algorithm for maximum flow velocity waveform estimation.

17

Table
‎
4
1

: Compliance with design constraints.

25

GLOSSARY

Abbreviation

Description

AC

Alternating Current

DC

Direct Current

DSP

Digital Signal Processing

EED

Electrical Engineering Department

PWM

Pulse Width Modulation

RMS

Root Mean Square

CVD

Cardiovascular disease

ML

Machine Learning

DL

Deep Learning

SNR

Signal-to-Noise-Ratio

UHF

Ultra-High-Frequency

IEEE

Institute of Electrical and Electronics Engineers

CCA

Common Carotid Artery

TIA

Transient Ischaemic Attack

ICA

Internal Carotid Artery

CAS

Carotid Artery Stenosis

DSA

Digital Subtraction Angiography

MRI

Magnetic Resonance Imaging

PW

Pulse Wave

CW

Continuous Wave

TCD

Transcranial Doppler Ultrasonography

PDU

Pulsed Doppler Ultrasonography

IQ

In-phase and quadrature components

TBI

Traumatic Brain Injury

MCA

Middle Cerebral Artery

ICU

Intensive Care Unit

IMT

Intima-Media Thickness

SAE

Stacked Autoencoders

Introduction

The purpose of this chapter is to provide a background and definition of the problem, which illustrates the design of wearable neckband for early detection of carotid artery stenosis. In addition, the aims and objectives will be highlighted in this chapter.

Background

A major cause of death in metropolitan cities is cardiovascular disease (CVD). In 2008, up to 17.3 million people died from cardiovascular disease, according to a World Health Organization survey. As predicted by scientists, nearly 23.3 million people will die because of cardiovascular diseases by 2030 ‎[1]. For the investigation of cerebrovascular and cardiovascular disorders, non-invasive carotid artery blood flow monitoring is necessary ‎[2]. Blood vessels that deliver blood to the brain, face, and neck are known as carotid arteries. The human body contains two common carotid arteries, one on each side of the neck. These common carotid arteries extend from the chest up to the skull. Each artery among them splits into two branches, the internal carotid artery, and the external carotid artery. As a result of the splitting of internal and external carotid arteries, several smaller artery branches nourish the tissues and organs of the head and neck. The carotid arteries play an important role in the circulatory system. Their purpose is to carry oxygen-rich blood to organs and tissues in the head and neck, including those in the brain. In an ideal world, this would be an easy journey. However, a blocked carotid artery or blood clot can make this process difficult and can cause serious medical problems. What has been described so far is illustrated in Figure 1.

Figure 1: Anatomy of Carotid Artery‎[3].

Machine Learning (ML) and Deep Learning (DL) are ground-breaking fields in the field of computer science. An approach to teaching computers and machines to learn from past data to predict future data or behaviour is known as machine learning. Unsupervised data is trained and learned using artificial neural network techniques and algorithms in deep learning, a subfield of machine learning ‎[4]. As an example, the figure shown below outlines the working principle of machine learning and deep learning ‎[5].Using the machine learning model, images are taken behaviour and their features are extracted. The model predetermines the output and applies the classification algorithm. However, in deep learning, the images are fed directly to the algorithms, and this does not require any manual feature extraction. The pictures pass to the several layers of the artificial neural network and anticipate the result ‎[6].

Figure 2: Machine Learning vs Deep Learning ‎[5].

In recent years, medical professionals have made extensive use of machine learning in biomedical applications to diagnose patients quickly and accurately. It can be used to predict the severity of a disease and plan treatment or therapy accordingly. Biomedical signals are mostly electrical signals caused by the electrochemical process of certain cells in the human body. A person’s health information can be gathered by measuring these signals. The useful signal is tainted with a variety of noise and unwanted artifacts when these signals are captured with an appropriate electrode. Biomedical signal processing involves analysing measured data to gather relevant information. Sifting, highlight extraction and classification of signals are needed to obtain important data. Typically, biomedical signals were captured and plotted through programming to support medical professionals with effective data. This helps them in understanding and decide a patient’s health status ‎[7].

Problem Definition

Without early diagnosis, stroke is the leading disabling disease. After cancer and heart disease, the third most common cause of death ‎[8]. In addition to causing a stroke, carotid artery disease can cause other cardiovascular conditions. As cholesterol and fatty material accumulate on arterial walls as shown in figure 2, it is identified as a chronic degenerative disease that remains asymptomatic. Thus, carotid artery stenosis causes arterial walls to become thicker and less elastic. Strokes are caused by plaques blocking blood vessels, which can remain undetected for a long time. It is therefore crucial to diagnose carotid artery stenosis at an early stage.

Figure 3: Carotid Artery Stenosis ‎[9].

Aims and Objectives

·
Aim: To detect carotid artery stenosis when it is at its early stages by designing a wearable neckband that monitors the health status of patients.

·
Objective: To achieve the aim, the following objectives have been specified:

1- Designing a signal acquisition system.

2- Acquisition, pre-processing, and digitization of doppler signal.

3- Transmitting the signal to the cloud server.

4- Utilizing open-source data set to train a deep learning model.

5- Deploying the deep learning model in the cloud server for real-time detection.

Design Constraints and Standards

Before starting any project, there are certain design constraints and standards that must be defined. In this project, a list of system constraints and standards are summarized in this section.

Table ‎11: List of the system constraints.

Specification

Constraints

Value

Economic

The system should be designed within a specified budget

7000QR

Technical

· Communication protocol – Bluetooth low energy to minimize power consumption
· SNR – A higher SNR value indicates that the signal is of high quality
· Signal Bandwidth (Of doppler signal)
· Sampling frequency – To prevent aliasing, a sampling frequency that is 2*fmax is required [18]
· Supply voltage
· Accuracy of deep learning model
· Portability
· Model decision time

· __
· >=20 dB

– 0.1 to 20 Hz
– 100 Hz

· 5V
· >= 90%
· Battery operated
· <10 ms Comfortability The device is lightweight therefore, it is comfortable on the patient’s neck 350g Life cycle It has a long-life cycle It’s a rechargeable device Time Time allocated to finish the project 8 months Safety The leakage current constraint on the human neck <= 100 Microamps

Table ‎12: List of the system standards.

Standard Description

Bluetooth

Utilizing UHF (Ultra-High Frequency) radio waves at a frequency between 2.4 and
2.485 GHz, this technique uses small distances to transfer data between devices ‎[10]

Medical electrical equipment

IEC 60601-1:2022 SER Series ‎[11]

Ultrasound machine

IEEE 790- 1989:
This standard outlines the methods for determining the pressure, power, and intensity of the ultrasound field ‎[12]

Project Plan

This section provides a detailed timeline of who will accomplish the project’s tasks and when they will be finished.

#

Task

Responsibility

Weeks

1

2

3

4

5

6

7

8

9

10

11

12

13

14

1

Background & Problem definition

2

Standards and limitations

3

Aims and objectives

4

Design constraints & standards

5

Report organization

6

Literature review

7

System Design

8

Design testing & validation

9

Conclusion & Future Work

Report Organization

This report consists of five main chapters. The first chapter is an introduction which includes a background and the project’s problem definition, followed by the aim of the project and the objectives. The chapter ends with the design constraints and standards, in addition to the project plan. The second chapter is a literature review about the detection techniques and more theory about Carotid Artery Stenosis disease. Next will be chapter three which is about the system design. In this chapter, a full system diagram is shown followed by an explanation of the hardware design description. The fourth chapter is to show the results obtained. Finally, the fifth chapter which concludes the project by summarising the most important points.

Design and Implementation of a Wearable Carotid Neckband Doppler System for Early Detection of Carotid Artery Stenosis 5

Literature Review

Carotid Artery Stenosis

The number of people with clinically severe stenosis in their carotid arteries varies by racial group. Native Americans and Caucasians seem to have the highest prevalence of carotid artery stenosis, whereas Asian women and African American men seem to have the lowest prevalence. According to this information, African American participants are likely not more susceptible to strokes due to extracranial cerebrovascular disease ‎[13]
. Atherosclerosis occurs when the inner wall of an artery becomes damaged. As the plaque accumulates in the body, it settles in the walls of the arteries. Among the substances that make them up are cholesterol, fibrous tissue, and calcium. There are several factors contributing to plaque formation, including age, smoking, diabetes, high cholesterol, obesity, inadequate exercise, and hereditary ‎[14]. The internal and external carotid arteries, which supply the brain and face, branch from the left and right Common Carotid Arteries (CCA) and travel up the neck. Stenosis often develops distally to the bifurcation of the CCA due to regional fluid dynamics. The Internal Carotid Artery (ICA), which supplies blood to the brain Transient Ischemic Attack (TIA), has a limited blood flow, making strokes more likely ‎[15]. Around 7.5% of men and 5.0% of women over 80 have carotid artery disease; the risk increases with age. As a consequent of carotid disease, plaque rupture and subsequent brain embolism are the primary causes of cerebral ischemia. Numerous factors influence the likelihood of cerebrovascular events brought on by carotid stenosis. Overall, patients with symptomatic carotid stenosis have a higher risk of stroke than patients with asymptomatic carotid stenosis. The first several weeks following the presenting event are when stroke risk is at its peak. People who had a hemispheric stroke had a higher probability of recurrence than patients who had a Transient Ischemic Attack (TIA) ‎[16]. To prevent stroke that could arise from carotid artery stenosis in recently diagnosed patients, the most efficacious technique is carotid endarterectomy. Additionally, as long as the risk of stroke and death following surgery is less than 3%, it helps patients who are younger than 75 years of age and have not yet developed signs of stroke ‎[14].

Detection Techniques

Imaging has advanced to a high level of complexity over the last few decades. The field of imaging has undergone a revolution with the advancement of 1D to 2D and 2D to 3D images. This not only aids in early detection of several serious and deadly diseases, but it also helps physicians make wise clinical judgments regarding the course of subsequent therapy. Accurate early diagnosis of Carotid Artery Stenosis (CAS), which has the potential to result in a crippling stroke, is crucial. There are several detection techniques in medicine field such as: X ray imaging, Digital Subtraction Angiography (DSA), Magnetic Resonance Imaging (MRI), and Duplex Ultrasound ‎[17].

X-Ray Imaging

Rontgen first used X-rays to image biological tissues in 1895, which is when X-ray imaging began. When radiation travels through different tissues, its intensity changes because of attenuation and scattering (different optical properties). When it hits a photographic film, it produces a 2D image. Bone (high attenuation for X-ray photons) imaging was the first discipline of medicine to use this technology ‎[13]. X-ray imaging necessitates the injection of a contrast agent to show the blood flow in the arteries. Contrast agents that attenuate X-rays, such as iodine, lanthanide, gold nanoparticles, etc., are assessed as contrast density values in Hounsfield units (Hu). The procedure is known as X-ray angiography.

Digital Subtraction Angiography (DSA)

Digital Subtraction Angiography (DSA) is a new technology that shares a lot of traits with the CT scanner that was created in the 1970s. DA is a computer-assisted method that, like CT scanning, combines digital data collecting with computer processing to create a medical image ‎[18]. DSA has sensitivity, specificity, and accuracy of 95%, 99%, and 97%, respectively ‎[17].

Magnetic Resonance Imaging (MRI)

Over the past 20 years, the technique of Magnetic Resonance Imaging (MRI) has undergone constant development, leading to MR systems with higher static magnetic fields, faster and stronger gradient magnetic fields, and more potent radiofrequency transmission coils. Due to the emergence of various new indications over the past few years, such as cardiovascular MRI, it is becoming more and more popular and in demand ‎[19]. Up until recently, gradient-based imaging technology improvements were the main strategy for accelerating MRI imaging speed. By doing this, it has been possible to apply magnetic field gradient pulses to a sample more quickly and with greater power. However, the FDA has put restrictions on high-speed gradient imaging techniques, limiting the patient’s exposure to Radiofrequency (RF) radiation and magnetic field gradients to prevent bioeffects ‎[20].

Duplex Ultrasound

In the 1950s, Doppler Ultrasound was used for the first time in medicine to assess the speed of heart walls and valves. The first duplex apparatus, featuring both Doppler and B-mode imaging, was described by Barber et al. in 1974. A doppler works on the principle that clinical ultrasound equipment produces images by using substantially the same concepts of other devices. Thus, 2 MHz to 10 MHz of sound are transmitted into the body and echoes from variations in acoustic impedance are detected. The variations in impedance are typically correlated with tissue interfaces, such as the junction between muscles and fat. A “duplex” transaction involves two modes of ultrasonic communication: doppler and b-mode. The B-mode transducer is a device which operates similarly to a microphone, in that it uses high-frequency waves to generate a picture of the blood vessel under examination. The two-dimensional cross-sectional images of the human body can be created using the non-invasive B-mode, or brightness-mode ultrasound, technology. Ultrasound pulses are used to calculate the depth of tissue contact, similar to the function of A-mode imagers. However, in B-mode imaging, the amplitude of the received echo is used to adjust the brightness of the image on a two-dimensional display ‎[20]. The B-mode and colour flow doppler images, as well as the duplex ultrasound signals, are shown in Figure 3.

Figure 4: Duplex Ultrasound ‎[21].

After a brief explanation of the different detection modes followed by an explanation of their history, function, and application. Doppler Ultrasound is used by medical professionals to identify cardiovascular (heart and blood vessel) issues. It measures the speed and direction of blood flow through arteries and veins. In addition to detecting blood clots, restricted arteries, and other issues with the heart and blood vessels, it can also detect problems with the legs, arms, and stomach. Vascular dopplers will be discussed in the next section.

Vascular Doppler

In the previous 20 years, Vascular Dopplers have not much improved. The audio output from modern gadgets must be carefully interpreted, which is frequently subjective, and they are big, expensive, and complicated. Vascular Doppler’s prospective applications are constrained by these drawbacks, which make them challenging to use. Every heart cycle causes a change in the frequency spectrum of the blood flow velocity in the arteries; therefore, a blood flow Doppler signal is regarded as a cycle-stationary stochastic gaussian signal and is deemed as quasi-stationary signal in short segments (20-20ms). The Doppler detector probe described by Sotomura is the most basic one for measuring blood flow. These products of probes are currently employed more frequently as diagnostic instruments for cardiovascular illnesses since they are more efficient and smaller. As illustrated in Figure 4, the probe consists of several basic components ‎[22].

Vascular Doppler is a tool frequently employed to evaluate clinically relevant variations in blood flow. A pulse wave (PW) or continuous wave (CW) ultrasound signal is delivered into a blood vessel and converted to audio by the reflected, Doppler-shifted sound wave. This enables the doctor to “hear” the patient’s blood flow. It aids in diagnosing diseases like arterial occlusion, arterial stenosis, embolism, thrombosis, arterial dissection, and scarring from repeated injections in an artery ‎[22].

Figure 5: Ultrasonic transducer (probe), PVC container, connector cable, and printed circuit board for a detector ‎[22].

Transcranial Doppler Ultrasound

Ultrasound is a safe diagnosis technique used in medical examinations owing to the application of short bursts of low-energy sound waves. Ultrasound imaging is most used for investigating the functioning of heart valves and for monitoring fetal health in pregnant women. It is the pulsed nature of blood flow that necessitates the use of pulsed Doppler ultrasound to detect arterial blood flow velocity ‎[23]. Doppler ultrasound is widely utilized for the identification of several neuro-critical disorders ‎[24]‎[25] due to its low cost and lack of invasiveness. Blood flow velocity changes are usually used in the diagnostic process ‎[24]‎[26]. When measuring blood flow from the basal arteries, transcranial Doppler (TCD) ultrasonography is a form of imaging that is quite comparable to Doppler ultrasound ‎[27]. Blood flow velocity profile changes can be used to detect arterial stenosis and occlusion, two common neurological disorders. The quality of the recorded TCD signal, however, depends heavily on the expertise of the operator, and as a result, TCD signals are frequently contaminated by motion artifacts and speckle noise ‎[28]‎[29]. An effective signal processing technique is required for detection of the change. As a result of its efficacy and dependability, computer-aided detection and classification systems have recently emerged as the norm ‎[30].

Pulsed Doppler Ultrasonography (PDU) is a technique that is frequently utilized in order to get measurements of the speed of blood flow ‎[23]. In PDU, short acoustic pulses that have a set carrier frequency, fc are transmitted toward a target segment of a blood artery. These pulses are directed at the area of interest. Within the insonated target volume, an ensemble of scatterers (red blood cells) moving at varying velocities reflects the acoustic pulses that are sent out. Reflection of acoustic pulses from moving red blood cells causes unique frequency shifts in the ultrasound echo signal. These shifts can be used to diagnose a variety of medical conditions.

The in-phase and quadrature components of the received echo signals are separated, and their separation is expressed by the (complex-valued) time-domain signal, IQ. The received echo signal is sampled, digitized, demodulated, and divided in this process. Spectral analysis of the IQ signal is then used to derive the time-varying frequency content of the demodulated echo signal and, as a result, the velocity distribution of the scatterer ensemble. The time-frequency spectrogram, also known as the time-velocity spectrogram, is denoted by the notation, SP, where the first argument is time and the second argument is frequency or, equivalently, velocity. This notation is shown in Figure 6 (Block C). The maximal flow velocity, which is the TCD quantity that has the most significant clinical relevance, is represented by the envelope of the Doppler spectrogram (see Figure 6 Block C for reference).

Figure 6: Doppler signal collection from middle cerebral artery (MCA) and Internal Carotid Artery (ICA).

Machine Learning

Most often, transcranial Doppler ultrasound is used to assess blood velocity in the middle cerebral artery (MCA) or internal carotid artery (ICA) to infer cerebral blood flow (CBF) ‎[31]. There is a risk of strokes due to carotid artery stenosis, hence the maximum flow velocity waveform can also be utilized to identify patients in critical conditions. Therefore, a classification of these critically ill patients during doppler diagnosis can aid in early diagnosis. With its ability to perform crucial classification jobs that can save lives at a moment’s notice, deep learning algorithms are finding success across the whole medical sector ‎[33]‎[37].

Stenosis Detection

This subsection gives an introduction on work previously done for carotid Intima-Media Thickness (IMT) segmentation.

Medical Encoder-Decoder Applications

Encoder-decoder models, which include merged techniques such as the stacked autoencoders (SAE), stacked denoising autoencoders (SDAE), stacked sparse autoencoders (SSAE), and convolutional variational autoencoders (CVAE), are currently gaining popularity in medical imaging. CVAE has been the subject of numerous studies, including medical applications for predicting health outcomes following trauma ‎[38]. In medical applications, encoder-decoder models have been the subject of several studies, including mortality risk prediction ‎[39] and chest radiology improvement with denoising autoencoders ‎[40]. Encoder-decoder models outperform other models when it comes to the accuracy of applications for medical image segmentation. As a result, the authors of ‎[41] asserted that their CT scan-based application for 3D image segmentation yields better results.

Carotid IMT Segmentation Applications

Intima-Media Thickness (IMT) segmentation is the most sensitive step because the thickness measurements depend on the accuracy of the IMT segmentation. Despite numerous attempts to segment and classify the carotid IMT, only a few achieve competitive results. The creators of [16] use support vector machines to prepare and fragment the carotid IMT. In their technique, they utilized 49 ultrasound pictures and partitioned them into two sets with half for preparing and the rest for testing. They found that the IMT measurement was 0.66 mm, and their method had an accuracy of 93%. Carotid segmentation was first done by Loizou et al. (2013) ‎[42], where they used a snake-based semi-automated segmentation system that was suitable for complete common carotid artery (CCA) segmentation. By manually defining the carotid plaque’s diameter and applying the snake algorithm to obtain measurements, their method concentrated on estimating IMT measurements. In addition, the authors of ‎[43], attempted to implement an adaptive snake’s contour and level-set segmentation based fully automated segmentation system. The authors discovered that when comparing the two methods together, the snake’s contour method performed better than level set segmentation. The authors of ‎[44] developed an additional method that avoided using deep learning for IMT segmentation. The authors demonstrated how they fully developed a fully automated region of interest (ROI) extraction as well as a threshold-based method for the intima-media complex and used a wind-driven optimization technique for carotid IMT segmentation. A screening tool that integrates a two-stage artificial intelligence model for IMT and carotid plaque measurements, made up of a Convolutional Neural Network (CNN), and a fully convolutional network (FCN), was used in experiments on IMT segmentation ‎[45]. The first deep learning model divides the CCA from the ultrasound image into two categories: rectangular wall patches and non-wall patches. The system then goes through two deep learning models. After that, the area of interest is looked at and sent to the second stage, where some features are found to figure out the carotid IMT and plaque total. The authors of ‎[46] proposed a CNN-based method for segmentation as deep learning and machine learning continued their studies of IMT segmentation. As a result, the researchers utilized an algorithm that makes use of the right-layer CNN architecture to locate the ROI. Additionally, they used 220 left and right CCA images to train the network for ROI localization. To measure the IMT, the intima-media complex area is then extracted. The IMT measurement yielded a mean difference of 0.08mm, with an accuracy of 89.99% for the CNN network. IMT segmentation in CNN-based video interpretation of IMT measurements was the subject of an investigation by another research group ‎[47]. They claimed that they were able to achieve a low error rate in the measurements with a result of 2.1mm error and only 1 failure for testing subjects when they carried out CNN with 6 layers. Besides, another procedure was utilized by Joseph and Sivaprakasam (2020)‎[48], where they utilized twofold line reverberation designs coming from the B-mode and A-mode ultrasound pictures to distinguish both blood vessel walls. Their method revealed an IMT measurement error of 0.18 mm. In ‎[49], researchers talked about another deep learning method that used CNN with multiple hidden layers to classify images. They were able to test the network with a dataset of 501 ultrasound images and achieve an IMT classification accuracy of 89.1%. The authors of ‎[50] devised an alternative strategy, which made use of 4 classification algorithms for IMT measurement: random forest, support vector machines (SVM) with radial basis kernel, and SWM with linear kernel.

System Design

Full System Block Diagram

According to the block diagram below, both conventional and proposed methods are represented in the full system. The proposed method involves developing and implementing a wearable carotid neckband. A neckband ultrasound will detect the patient’s health status, which will then be transmitted to a waistband device for modification and filtering, and then transmits signal data via Bluetooth to a mobile phone. In the first place, it will display the right and left common carotid arteries and then send it to the server to check healthy subjects and ICU patients. Lastly, the results will appear on the mobile phone.

Figure 7: Full system block diagram.

Neckband 3D model

The 3D model below shows the design of a wearable Carotid neckband that will be used for early detection of Carotid Artery Stenosis. This will be adjusted for different neck sizes. In addition, the design will be developed since it will be wired with the waistband.

Figure 8: Neckband 3D model.

Neckband and Waistband Electric design

For Neckband

For Waistband

Figure 9: Doppler Probe.

A doppler probe is composed of two parts that will be split. The first part is for the neckband that serves as the head of the device, measuring the patient’s position through his neck, and the second part is for waistband that will act as signal conditioning circuits to amplify, filter, and modify the received signal.

Machine Learning Model

From data pre-processing through classification model development, this study used several techniques to distinguish between signals from healthy subjects and those from ICU patients. The study’s methodology and resources are summarized in Figure 10. The following sections detail every individual stage of the overall system.

Dataset Description

A dataset of transcranial doppler ultrasound scans collected by ‎[26] was used in this study. In addition to data from healthy participants, the dataset also includes information from those with diagnosed neurocritical disorders. Six volunteers between the ages of 25 and 45 were used as healthy subject in the study ‎[26]. A total of sixteen, approximately two-hour-long recordings were made from the MCAs and ICAs of healthy volunteers. The 29 recordings were taken from twelve patients aged 23 to 74 years old who were admitted to neurocritical care for conditions such hydrocephalus, traumatic brain injury (TBI), and subarachnoid or intraparenchymal haemorrhage. A portable ultrasound equipment (Philips CX-50TM) was used in the data collection process which was equipped with an S5-1 ultrasonic transducer operating at 1.75 MHz with a 90 field of view to capture the blood flow velocities. The block A, B and C of the Figure 10 illustrates the data collection process adopted in ‎[26], where the ultrasound probe was positioned over the M1 segment of the middle cerebral artery and the temporal area to evaluate flow velocity ‎[26].

Figure 10: Overview of the method and materials used in this study to classify “Healthy” and “ICU” patients.

Signal Extraction

Using a short burst of ultrasonic signal with a 1.75 MHz carrier frequency (fc) directed towards the MCA/ICA region, the acoustic signal reflected by moving red blood cells (RBC) was used to generate the transcranial doppler (TCD) echo signal. A pipeline is proposed in ‎[26] to extract maximum flow velocity waveform where Short-Time Fourier Transform (STFT), envelope tracing, and post-processing were used. The STFT is used to calculate the spectrogram for the TCD echo signal. After that, an envelope tracing algorithm was applied to get maximum flow velocity waveform with the highest signal quality index. Algorithm 1 represent the envelope tracing algorithm proposed in ‎[26]. Noise and artifacts are common in maximum flow velocity waveform due to rapid oscillation. A post-processing technique using 4th-order Butterworth filter and 1D-median filter to counter the noise and artifacts was proposed ‎[26]. The maximum flow velocity waveform estimation steps are shown in Figure 10 (Blocks D through E), are integral to the overall approach taken in this investigation. The maximum flow velocity waveform taken from Middle Cerebral Artery is considered as “MCA/ICA waveform” for further mentions in the following sections.

Table ‎31: Algorithm Envelope tracing algorithm for maximum flow velocity waveform estimation.

Algorithm

Envelope tracing for maximum flow velocity waveform

Input:

Doppler Signal Spectrogram

Step 1:

Black & White Thresholding

Step 2:

2D Median Filter

Step 3:

Envelope Tracing

Step 4:

Signal Quality Index

Step 5:

Keeping Flow Velocity with Best Result

Step 6:

Check Maximum Iteration
if No:
Adapt Intensity Threshold return Step 1
If Yes:
return Maximum Flow Velocity Waveform

Manual Data Cleaning

After the maximum flow velocity waveform estimation, further data cleaning was needed to exclude corrupted signal. As the recordings were taken in long sessions, the data signal is segmented into small 1024 data segments for cleaning. The segmented signals with unwanted noise, artifacts and signal jumps in many instances were removed. For that, MATLAB app was developed, and Supplementary Figure 11 represents the MATLAB App in App Developer view.

Figure 11: Graphical presentation of high quality, low quality, mediocre and corrupted signals.

Depending on the quality of the signal, the segment was annotated as one of the four options- high, low, mediocre, and corrupted. Figure 11 illustrates the four types of signals available in the dataset. The trained human annotators were strict in choosing the high and low-quality segments, following the data preparation strategies proposed in ‎[51]. The signals labeled as high are uniform throughout the entire frame with stable signal and no baseline wandering. On the other hand, the signals labeled as low are detectable signals but contains baseline shifts, major and minor distortion due to motion artifacts. The third option, mediocre was not used for next step; rather, it would create ambiguity. Finally, the signal segments that cannot be identified as to what type of signal it may be, were labeled as corrupted. These segments were later discarded. Figure 12 illustrates the dataset details before and after manually labeling and cleaning. This dataset will be used in SDPII to train CycleGAN to improve the quality of the doppler signal and then will be used for training deep learning model.

Figure 12: Dataset details before and after manually labeling and cleaning.

Performance Metrics

The performance of the proposed models in this study was evaluated using evaluation metrics, such as overall accuracy, precision, recall, f1 score, and specificity, receiver operating characteristic (ROC) curves. As the sample number in Healthy and ICU patient classes are not exactly same, weighted metrics of precision, recall, f1 score and specificity will be calculated. Overall accuracy, weighted sensitivity or recall, specificity, precision, and F1 score are mathematically represented in Equations 1-5.

( 1 )

( 2 )

( 3 )

( 4 )

( 5 )

Here, α = True positive, β = False positive, γ = True negative, and δ = False negative.

Design Testing and Validation

Doppler Simulation Results

This section includes simulation results of the system to be developed. The system consists of several blocks to be simulated. The first step is to transmit a signal to the transmitter circuit. The signal transmitter adopts a continuous wave ultrasonic transducer with a working frequency of 2MHz. The transmitting circuit adopts a passive crystal oscillator, a 3-point oscillation circuit, and a gain amplifying circuit to form a signal-generating and transmitting circuit and provides a working signal for the signal amplifier. The power supply is used to provide working power for the transmitting circuit, demodulation circuit, filtering circuit, and the audio amplifier. In addition, the signal receiver is a continuous wave ultrasonic transducer which is connected to the demodulation circuit. The demodulation and signal receiving circuit carries out analog signal processing to the echo signal received by the ultrasonic transducer, and it adopts an analog multiplier chip to form the demodulation circuit. The filter circuit is connected to the demodulation circuit, and further processes the output signal of the demodulation circuit. The audio amplifier is connected to the output end of the filter circuit, and the audio signal is further amplified, and the output goes through the headphone jack. 

Crystal Oscillator Transmitter

A crystal oscillator is an electrical circuit that creates a sinusoidal electronic signal at a highly specific frequency using the mechanical resonance of a vibrating crystal. At this point, an ultrasonic continuous wave at the operation frequency (4, 5, 8, 10 MHz) is created. Additionally, the quadrature signals Sin (ω0t) and Cos (ω0t) are generated at the same transducer operation frequency during this stage ‎[22]. Ultrasonic transducers are based on piezoelectric materials’ converse and direct effects. When a potential difference is applied across the electrodes, a vibration occurs, and when an echo is received, a signal is produced. Consequently, transducer technology relies heavily on piezoelectric components. The right piezoelectric materials are selected for specific applications based on their piezoelectric performance, dielectric properties, elastic properties, and stability ‎[63]. The Crystal Oscillator Transmitter circuit is made of several components, such as: capacitors, resistors, transistor, inductor, and crystals. Multisim was used to implement the circuit which is shown in the figure below: 

Figure 13:Simulation of Crystal Oscillator Transmitter Circuit using Multisim.

Figure 14: The output waveform of the transmitter circuit.

A crystal oscillator’s output is an AC waveform with a closely regulated frequency. According to how the crystal is cut from the quartz blank, it typically has the appearance of a sine wave and may oscillate on its primary frequency or on a second, third, or higher overtone (harmonics). The oscillator circuit must first amplify the relatively little oscillating voltage before it can be used. 

Signal Receiver

The output voltage of the MC1496 balanced modulator-demodulator is a function of the input voltage signal and the switching function (carrier). Suppressed carrier and amplitude modulation, synchronous detection, FM detection, phase detection, and chopper applications are a few examples of typical uses.  This circuit consists of an upper quad differential amplifier driven by a standard dual current source differential amplifier. A sinewave carrier input signal of 60 mV rms was used to describe the MC1496. For balanced modulator applications, this level typically comes highly recommended since it offers the best carrier suppression at carrier frequencies close to 500 kHz. Signal level, VS, has no effect on carrier feedthrough. Thus, operating at high signal levels will maximise carrier suppression. The signal-input transistor pair must, however, remain in a linear operational mode to prevent harmonics of the modulating signal from being produced and appearing in the device output as erroneous sidebands of the suppressed carrier. This criterion caps the amplitude of the incoming signal. Also keep in mind that for good carrier suppression and minimal spurious sideband production, an ideal carrier level is advised. Circuit design is crucial for reducing carrier feedthrough at higher frequencies. To avoid capacitive coupling between the carrier input leads and the output leads, shielding may be required.

Figure 15: Balanced Modulator & Demodulator circuit connection.

Figure 16: Balanced Modulator and Demodulator circuit output.

Filter Circuit

A filter is a circuit that allows some frequencies to pass through (or be amplified) while attenuating others. Consequently, a filter can remove crucial frequencies from signals that also include unwanted or unrelated frequencies. The highest frequency components of interest in a signal correlate to the signal bandwidth (BW) theoretically. The bandwidth, which is used to describe the separation of various signals or the dynamic behaviour of a signal, is equivalent to the cut-off frequency. This is the case, for instance, when signals that are closely spaced in frequency need to be separated or when their amplitudes or phase values must change quickly, as in imaging applications. The figures below show the circuit connection of the filter that will be used, in addition to the obtained output. 

Figure 17: Simulation of Bandpass Filter circuit.

Figure 18: Bode plots of a Bandpass filter.

Audio Amplifier

As a basic circuit arrangement, audio amplifiers magnify audio signals coming in through microphones, going out via speakers, radios, wireless transmitters, etc.  Fundamental power amplifiers produce high-power output signals from low-power input sources. The different domains where an electrical signal is transformed into an auditory signal make use of this amplification process. Audio amplifiers are this kind of amplifier. The audio amplifier is present both at the input and the output of every circuit that handles audio signals. For instance, before further processing a sound wave input signal received by a microphone, the signal needs to be pre-amplified. Similarly, before delivering an electrical signal to a speaker, it needs to be amplified. The simulated circuit and the resultant output shown below, where the red wave represents the input, and the blue wave represents the output after amplifying. 

Figure 19: Audio Amplifier multisim circuit.

Figure 20: Output of audio amplifier circuit.

Power Supply Circuit

The power supply is made using an LED, filter capacitors, battery, switch, and transistor. The purpose of this circuit is to provide power to the transmitter, demodulation circuit, filter circuit and audio amplifier ‎[64]. The LED is used to indicate the power. When the switch is off, the battery gets disconnected and when it is on, a voltage is produced. This circuit is simulated using Multisim as shown below.

Figure 21: Power supply circuit simulation.

Compliance with Design Constraints

This section includes a table that summarizes the design constraints that were discussed previously and whether they have been met or not.

Table ‎41: Compliance with design constraints.

Specification

Constraint

Value

Actual System

Economic

The system should be designed within a specified budget

7000QR

–

Technical

· Communication protocol – Bluetooth low energy to minimize power consumption
· SNR – A higher SNR value indicates that the signal is of high quality
· Signal Bandwidth (Of doppler signal)
· Sampling frequency – To prevent aliasing, a sampling frequency that is 2*fmax is required [18]
· Supply voltage
· Accuracy of deep learning model
· Portability
· Model decision time

· __
· >=20 dB

· 0.1 to 20 Hz
· 100 Hz
· 5V
· >= 90%
· Battery operated
<10 ms - Comfortability The device is lightweight therefore, it is comfortable on the patient’s neck 350g - Life cycle It has a long-life cycle It’s a rechargeable device - Time Time allocated to finish the project 8 months - Safety The leakage current constraint on the human neck <= 100 Microamps -

As far as the design constraint is concerned, unless the implementation is not finalized it cannot be said that the requirements are met.

Conclusion & Future Work

Conclusion

In conclusion, a description of the problem was introduced in this report. A lightweight device used to detect carotid artery stenosis and monitor the health status of individuals without the need for assistance from physicians is the primary goal of this project. The literature review includes various detection techniques such as X-Ray, DSA, MRI, Duplex Ultrasound, Vascular Doppler, and Transcranial Doppler Ultrasound. In addition, a machine learning part that summarizes the different techniques used to classify patients as healthy or unhealthy is included. The system design chapter provides block diagrams with descriptions of the system functionality. Additionally, the simulation is used to depict the working principle of ultrasonic transducers. The device will use Bluetooth communication to send the signals to a cloud server, where the deep learning model will be deployed.

Recommendations for Future Work

For future work recommendations, having a deep knowledge in machine learning and deep learning is crucial especially in CIMT segmentation. A duplex doppler system will be used to get 2D image and doppler signal to calculate cIMT as ground truth for stenosis detection. The vascular doppler signal will be improved from the bad quality signal to a good one using CycleGAN model in SDP II. In addition, the hardware design will be implemented as a neckband and waistband and the deep learning model will be implemented and deployed in cloud server for real-time detection.

REFERENCES

V. Naik, R. Gamad and P. Bansod, “Carotid Artery Segmentation in Ultrasound Images and Measurement of Intima-Media Thickness”, BioMed Research International, vol. 2013, pp. 1-10, 2013. Available: 10.1155/2013/801962.

Song et al., “Design and Implementation of a New Wireless Carotid Neckband Doppler System with Wearable Ultrasound Sensors: Preliminary Results”, Applied Sciences, vol. 9, no. 11, p. 2202, 2019. Available: 10.3390/app9112202.

“What are the carotid arteries?” Cleveland Clinic. [Online]. Available: https://my.clevelandclinic.org/health/articles/21492-carotid-artery. [Accessed: 31-Dec-2022].

J.-É. S. Kenny, I. Barjaktarevic, A. M. Eibl, M. Parrotta, B. F. Long, and J. K. Eibl, “A wearable carotid Doppler tracks changes in the descending aorta and stroke volume induced by end-inspiratory and end-expiratory occlusion: A pilot study,” Health science reports, 05-Oct-2020. [Online]. Available: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7534617/. [Accessed: 31-Dec-2022].

S. M. K. Kant, V. B. Semwal, V. Bijalwan, and Crespo Rubén González, “Machine Learning and Deep Learning: A Comparative Review,” in Proceedings of Integrated Intelligence Enable Networks and Computing: IIENC 2020, Singapore: Springer, 2022.

Info@lawtomated.com, “A.I. Technical – machine learning vs. Deep Learning,” lawtomated, 02-Feb-2020. [Online]. Available: https://lawtomated.com/a-i-technical-machine-vs-deep-learning/. [Accessed: 31-Dec-2022].

K. K. Singh, M. Elhoseny, A. Singh, and A. A. Elngar, “3 – Machine learning for biomedical signal processing.,” in
Machine learning and the internet of medical things in Healthcare, Amsterdam: Academic Press, 2021.

S. SAVAŞ, N. TOPALOĞLU, Ö. KAZCI, and P. KOŞAR, “Comparison of deep learning models in carotid artery intima-media thickness ultrasound images: Caimtusnet,” Bilişim Teknolojileri Dergisi, vol. 15, no. 1, pp. 1–12, 2022.

“Carotid artery stenosis: Causes, symptoms and treatment,” Cleveland Clinic. [Online]. Available: https://my.clevelandclinic.org/health/diseases/16845-carotid-artery-disease-carotid-artery-stenosis. [Accessed: 31-Dec-2022].

“Wireless standards explained: 802.11ax, 4G and bluetooth: Netxl,” NetXL Blog. [Online]. Available: https://www.netxl.com/blog/networking/wireless-standards/. [Accessed: 10-Sep-2022].

“IEEE Guide for Medical Ultrasound Field Parameter Measurements,” in IEEE Std 790-1989, vol., no., pp.0_1-, 1990, doi: 10.1109/IEEESTD.1990.94606.

M. Meola, J. Ibeas, G. Lasalle and I. Petrucci, “Basics for performing a high-quality color Doppler sonography of the vascular access”, 2022.

Harrington, D. P., Boxt, L. M., & Murray, P. D. (1982). Digital subtraction angiography: Overview of technical principles.
American Journal of Roentgenology,
139(4), 781–786. https://doi.org/10.2214/ajr.139.4.781

“The facts about carotid artery disease,” Baton Rouge Vascular Specialty Center, 17-Mar-2015. [Online]. Available: https://brvsc.com/the-facts-about-carotid-artery-disease/. [Accessed: 10-Oct-2022].

“Deep predictive learning of carotid stenosis severity.” [Online]. Available: https://www.researchgate.net/publication/353655573_Deep_Predictive_Learning_of_Carotid_Stenosis_Severity. [Accessed: 10-Oct-2022]. https://arxiv.org/abs/2108.00296

C. B. Rockman, H. Hoang, Y. Guo, T. S. Maldonado, G. R. Jacobowitz, T. Talishinskiy, T. S. Riles, and J. S. Berger, “The prevalence of carotid artery stenosis varies significantly by race,”
Journal of Vascular Surgery, vol. 57, no. 2, pp. 327–337, 2013. https://pubmed.ncbi.nlm.nih.gov/23177534/#:~:text=Conclusions%3A%20The%20prevalence%20of%20clinically,to%20have%20the%20lowest%20prevalence.

Saxena, A., Ng, E.Y.K. & Lim, S.T. Imaging modalities to diagnose carotid artery stenosis: progress and prospect. 
BioMed Eng OnLine 
18, 66 (2019). https://doi.org/10.1186/s12938-019-0685-7

Dill, T. (2008). Contraindications to magnetic resonance imaging.
Heart,
94(7), 943–948. https://doi.org/10.1136/hrt.2007.125039

GERALD BROWN, D. (2005, December). Instrumentation for parallel magnetic resonance imaging. 
INSTRUMENTATION FOR PARALLEL MAGNETIC RESONANCE IMAGING. https://oaktrust.library.tamu.edu/bitstream/handle/1969.1/4784/etd-tamu-2005C-ELEN-Brown ?sequence=1&isAllowed=y

W Rickey, D. (1991, March). Calibration of Doppler Ultrasound Instrumentation.
Calibration of Doppler Ultrasound Instrumentation. https://doi.org/10.13140/RG.2.2.11927.93608

R. M.- Hyperarts, “Carotid ultrasound,”
Vascular & Endovascular Surgery – Carotid Ultrasound. [Online]. Available: https://vascularsurgery.ucsf.edu/conditions–procedures/carotid-ultrasound.aspx. [Accessed: 01-Jan-2023].

Fuentes, M., Sotomayor, A., García, F., Moreno, E., & Acevedo, P. (2006). Design and construction of a blood flow detector probe for a continuous wave bidireccional Doppler ultrasound system.
Ingeniería Investigación y Tecnología,
7(2), 97–103. https://doi.org/10.22201/fi.25940732e.2006.07n2.008

Shung, K.K., Diagnostic ultrasound: Imaging and blood flow measurements. 2005: CRC press.

Christou, I., et al., A broad diagnostic battery for bedside transcranial Doppler to detect flow changes with internal carotid artery stenosis or occlusion. 2001. 11(3): p. 236-242.

Thorpe, S.G., et al., Velocity curvature index: a novel diagnostic biomarker for large vessel occlusion. 2019. 10(5): p. 475-484.

Wadehn, F., et al., Adaptive Maximal Blood Flow Velocity Estimation From Transcranial Doppler Echos. 2020. 8: p. 1-11.

A Carotid Doppler Patch Accurately Tracks Stroke Volume Changes During a Preload-Modifying Maneuver in Healthy Volunteers

Thorpe, S.G., et al., Toward automated classification of pathological transcranial Doppler waveform morphology via spectral clustering. 2020. 15(2): p. e0228642.

Aaslid, R., et al., Noninvasive transcranial Doppler ultrasound recording of flow velocity in basal cerebral arteries. 1982. 57(6): p. 769-774.

Sharma, S., et al., The role of transcranial Doppler in cerebral vasospasm: a literature review. 2020: p. 201- 205.

Jarrett, C.L., et al., Imaging transcranial Doppler ultrasound to measure middle cerebral artery blood flow: the importance of measuring vessel diameter. 2020. 319(1): p. R33-R42.

Kiranyaz, S., et al., Blind ECG Restoration by Operational Cycle-GANs. 2022.

Rahman, T., et al., QCovSML: A reliable COVID-19 detection system using CBC biomarkers by a stacking machine learning model. 2022. 143: p. 105284.

Chowdhury, M.H., et al., Lightweight End-to-End Deep Learning Solution for Estimating the Respiration Rate from Photoplethysmogram Signal. 2022. 9(10): p. 558.

Rahman, T., et al., COV-ECGNET: COVID-19 detection using ECG trace images with deep convolutional neural network. 2022. 10(1): p. 1-16.

Qiblawey, Y., et al., Detection and severity classification of COVID-19 in CT images using deep learning. Diagnostics. 2021; 11: 893. 2021, s Note: MDPI stays neutral with regard to jurisdictional claims in published.

Abir, F.F., et al., PCovNet: A presymptomatic COVID-19 detection framework using deep learning model using wearables data. 2022. 147: p. 105682.

Cakmak, A.S.; Thigpen, N.; Honke, G.; Alday, E.P.; Rad, A.B.; Adaimi, R.; Chang, C.J.; Li, Q.; Gupta, P.; Neylan, T.; et al. Using Convolutional Variational Autoencoders to Predict Post-Trauma Health Outcomes from Actigraphy Data. arXiv 2020, arXiv:cs.LG/2011.07406.

Alhassan, Z.; Budgen, D.; Alshammari, R.; Daghstani, T.; McGough, A.S.; Al Moubayed, N. Stacked Denoising Autoencoders for Mortality Risk Prediction Using Imbalanced Clinical Data. In Proceedings of the 2018 17th IEEE International Conference on Machine Learning and Applications (ICMLA), Orlando, FL, USA, 17–20 December 2018; pp. 541–546

Lee, D.; Choi, S.; Kim, H.J. Performance evaluation of image denoising developed using convolutional denoising autoencoders in chest radiography. Nucl. Instrum. Methods Phys. Res. Sect. A 2018, 884, 97–104.

Sital, C.; Brosch, T.; Tio, D.; Raaijmakers, A.; Weese, J. 3D medical image segmentation with labeled and unlabeled data using autoencoders at the example of liver segmentation in CT images. arXiv 2020, arXiv:eess.IV/2003.07923.

Loizou, C.; Kasparis, T.; Spyrou, C.; Pantzaris, M. Integrated system for the complete segmentation of the common carotid artery bifurcation in ultrasound images. In IFIP International Conference on Artificial Intelligence Applications and Innovations; Springer: Berlin/Heidelberg, Germany, 2013; Volume 412, pp. 292–301.

Christodoulou, L.; Loizou, C.P.; Spyrou, C.; Kasparis, T.; Pantziaris, M. Full-automated system for the segmentation of the common carotid artery in ultrasound images. In Proceedings of the 2012 5th International Symposium on Communications, Control and Signal Processing, Rome, Italy, 2–4 May 2012; pp. 1–6.

Madipalli, P.; Kotta, S.; Dadi, H.; Nagaraj, Y.; Asha, C.S.; Narasimhadhan, A.V. Automatic Segmentation of Intima Media Complex in Common Carotid Artery using Adaptive Wind Driven Optimization. In Proceedings of the 2018 Twenty Fourth National Conference on Communications (NCC), Hyderbad, India, 25–28 February 2018; pp. 1–6

Biswas, M.; Saba, L.; Chakrabartty, S.; Khanna, N.N.; Song, H.; Suri, H.S.; Sfikakis, P.P.; Mavrogeni, S.; Viskovic, K.; Laird, J.R.; et al. Two-stage artificial intelligence model for jointly measurement of atherosclerotic wall thickness and plaque burden in carotid ultrasound: A screening tool for cardiovascular/stroke risk assessment. Comput. Biol. Med. 2020, 123, 103847.

Madian, N.; Sunder, T. Convolutional Neural Network for Segmentation and Measurement of Intima Media Thickness. J. Med. Syst. 2018, 42.

Shin, J.Y.; Tajbakhsh, N.; Hurst, R.T.; Kendall, C.B.; Liang, J. Automating Carotid Intima-Media Thickness Video Interpretation with Convolutional Neural Networks. In Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV, USA, 27–30 June 2016; pp. 2526–2535.

Raj, K.V.; Joseph, J.; Sivaprakasam, M. Automated Measurement of Compression-Decompression in Arterial Diameter and Wall Thickness by Image-Free Ultrasound. Comput. Methods Programs Biomed. 2020, 194, 105557.

Sava¸s, S.; Topaloglu, N.; Kazcı, Ö.; Ko¸sar, P. Classification of Carotid Artery Intima Media Thickness Ultrasound Images with Deep Learning. J. Med. Syst. 2019, 43.

Qian, C.; Yang, X. An Integrated Method For Atherosclerotic Carotid Plaque Segmentation In Ultrasound Image. Comput. Methods Programs Biomed. 2017, 153.

Kiranyaz, S., et al., Blind ECG Restoration by Operational Cycle-GANs. 2022.

Park, Y.W., et al., Cycle-consistent adversarial networks improves generalizability of radiomics model in grading meningiomas on external validation. 2022. 12(1): p. 1-9.

Ganokratanaa, T., et al., Unsupervised anomaly detection and localization based on deep spatiotemporal translation network. 2020. 8: p. 50312-50329.

Ali, M.B., et al., Domain mapping and deep learning from multiple MRI clinical datasets for prediction of molecular subtypes in low grade gliomas. 2020. 10(7): p. 463.

Saqlain, A.S., et al. SL-CycleGAN: Blind Motion Deblurring in Cycles using Sparse Learning. in 2022 7th International Conference on Cloud Computing and Big Data Analytics (ICCCBDA). 2022. IEEE.

Keleş, O., et al. Self-Organized Residual Blocks For Image Super-Resolution. in 2021 IEEE International Conference on Image Processing (ICIP). 2021. IEEE.

Kiranyaz, S., et al., Operational neural networks. 2020. 32(11): p. 6645-6668.

Kiranyaz, S., et al., Self-organized operational neural networks with generative neurons. 2021. 140: p. 294- 308.

Malik, J., et al., Convolutional versus Self-Organized Operational Neural Networks for Real-World Blind Image Denoising. 2021

Yílmaz, M.A., et al. Self-Organized Variational Autoencoders (Self-Vae) For Learned Image Compression. in 2021 IEEE International Conference on Image Processing (ICIP). 2021. IEEE

Malik, J., et al., Self-organized operational neural networks for severe image restoration problems. 2021. 135: p. 201-211.

Sandler, M., et al. Mobilenetv2: Inverted residuals and linear bottlenecks. in Proceedings of the IEEE conference on computer vision and pattern recognition. 2018.

Q. Zhou, K. H. Lam, H. Zheng, W. Qiu, and K. K. Shung, “Piezoelectric single crystal ultrasonic transducers for biomedical applications,” Progress in Materials Science, vol. 66, pp. 87–111, 2014. 

杨进, 杨城, 郑超然, and 余圣越, “A kind of portable simple Doppler ultrasonography on monitoring instrument,” 03-Dec-2014. 

Li, Z., et al., Stenosis diagnosis based on peripheral arterial and artificial neural network. 2021. 10(1): p. 1- 21.

Appendix A CONTRIBUTION

Student Name

Student ID

Section Contributed to

1.1, 1.2,1.4,1.5
2.1,2.3
3.4,3.5
5.1, 5.2

1.3, 1.4
2.2
3.1,3.2,3.3

1.4,1.6

2.4

2.5

4.1

image1

image2

image3

image4

image5

image6

image7

image8

image9

image10

image11

image12

image13

image14

image15

image16

image17

image18

image19

image20

image21

image22

image23

image24

image25