Final Report Bme Design 3 Title Development Of Medical Devices.pdf

<span class="text_page_counter">Trang 1</span><div class="page_container" data-page="1">

SCHOOL OF ELECTRICAL & ELECTRONIC ENGINEERING

<b>ELECTRONICS DEPARTMENT</b>

<b>ADVANCED BIOMEDICAL ENGINEERING PROGRAM</b>

<b>FINAL REPORT</b>

<i><b>BME DESIGN 3</b></i>

<i><b>TITLE DEVELOPMENT OF MEDICAL DEVICES</b></i>

MENTOR

</div><span class="text_page_counter">Trang 2</span><div class="page_container" data-page="2">

HANOI, 2-2022

</div><span class="text_page_counter">Trang 3</span><div class="page_container" data-page="3">

<b>STUDENT INFORMATION</b>

<b>Student 1: </b>

Student ID:Class:Email:Cellphone:Avatar:

<b>Student 2: </b>

Student ID:Class:Email:Cellphone:Avatar:

<b>Student 3: </b>

Student ID:

</div><span class="text_page_counter">Trang 4</span><div class="page_container" data-page="4">

Class:Email:Cellphone:Avatar:

</div><span class="text_page_counter">Trang 5</span><div class="page_container" data-page="5">

<b> </b>

<b>EVALUATION OF REPORT BME DESIGN 3</b>

(For Instructor)The evaluated teacher’s name:

<b>Student's full nameStudent ID</b>

Project name:

<b>Choose the appropriate grades for students to present according to the followingcriteria: </b>

Very poor (1); Poor (2); Good (3); Very Good (4); Excellent (5)

<b>There is a combination of theory and practice (20)</b>

1 _{issues and hypotheses (including purpose and relevance) as}^{Clearly state the urgency and importance of the topic,}well as the scope of application of the project

<b>Be able to analyze and evaluate results (15)</b>

A clear work plan includes objectives and methods ofimplementation based on the results of systematic theoreticalresearch

1 2 3 4 5

6 ^{The results are presented logically and understandably, all}

results are analyzed and evaluated satisfactorily. ^{1 2 3 4 5}7 In the conclusion, the author points out the difference (if 1 2 3 4 5

</div><span class="text_page_counter">Trang 6</span><div class="page_container" data-page="6">

any) between the results achieved and the original goal andprovides an argument to suggest a possible solution in thefuture.

<b>Project book writing skills (10)</b>

The project presents the correct template with a logical andbeautiful chapter structure (table, clear image, titled,numbered order and explained or mentioned in the project,with margins, distances after dots, commas, etc.), withchapter opening and chapter conclusion, list of references andhave quotes in accordance with regulations

1 2 3 4 5

Excellent writing skills (standard sentence structure,scientific style, logical and well-founded reasoning,appropriate use vocabulary, etc.)

1 2 3 4 5

<b>Scientific research achievements (5) (choose 1 of 3 cases)</b>

Having scientific articles published or accepted forpublication/winning the 3rd prize of science at the Institute orhigher/scientific awards (international/domestic) from 3rdprize or higher / Patent registration

Reported at the Institute-level council during the scientificresearch student conference but failed to win the 3rd or higherprize/Won the consolation prize in other national andinternational examinations in majors such as TI contest.

<b>Total conversion score on scale 10</b>

<i><b>Other comments </b></i>(About the attitude and work spirit of the students)

</div><span class="text_page_counter">Trang 8</span><div class="page_container" data-page="8">

<b>EVALUATION OF BME DESIGN 3</b>

(For Reviewers)

The evaluated teacher’s name:...

<b>Student's full nameStudent ID</b>

Project name:

<b>Choose the appropriate grades for students to present according to the followingcriteria: </b>

Very poor (1); Poor (2); Good (3); Very Good (4); Excellent (5)

<b>There is a combination of theory and practice (20)</b>

1 _{issues and hypotheses (including purpose and relevance) as}^{Clearly state the urgency and importance of the topic,}well as the scope of application of the project

1 2 3 4 5

6 ^{The results are presented logically and understandably, all}

results are analyzed and evaluated satisfactorily. ^{1 2 3 4 5}

In the conclusion, the author points out the difference (ifany) between the results achieved and the original goal andprovides an argument to suggest a possible solution in thefuture.

1 2 3 4 5

</div><span class="text_page_counter">Trang 9</span><div class="page_container" data-page="9">

<b>Project book writing skills (10)</b>

The project presents the correct template with a logical andbeautiful chapter structure (table, clear image, titled,numbered order and explained or mentioned in the project,with margins, distances after dots, commas, etc.), withchapter opening and chapter conclusion, list of references andhave quotes in accordance with regulations

1 2 3 4 5

Excellent writing skills (standard sentence structure,scientific style, logical and well-founded reasoning,appropriate use vocabulary, etc.)

1 2 3 4 5

<b>Scientific research achievements (5) (choose 1 of 3 cases)</b>

Having scientific articles published or accepted forpublication/winning the 3rd prize of science at the Institute orhigher/scientific awards (international/domestic) from 3rdprize or higher / Patent registration

Reported at the Institute-level council during the scientificresearch student conference but failed to win the 3rd or higherprize/Won the consolation prize in other national andinternational examinations in majors such as TI contest.

<b>Total conversion score on scale 10</b>

<i><b>Other comments </b></i>(About the attitude and work spirit of the students)

</div><span class="text_page_counter">Trang 11</span><div class="page_container" data-page="11">

nghiê ]m cho em trong suốt 5 năm đại học vừa qua.Đ_ đ` tài này được hoàn thiê ]n thì emxin gHi lời cMm ơn chân thành đến thầy giáo ThS. Hoàng Quang Huy, giMng viênTrường Điê ]n - Điê ]n tH, Đại học Bách Khoa Hà Nội – là giMng viên hướng djn đồ ánnày của em, đ^ hết lòng giúp đỡ, hướng djn, chỉ dạy tận tình và động viên đ_ em hồnthành được đ` tài này.

Cpng với đó, em xin cMm ơn gia đình, bạn bè đ^ luôn bên cạnh giúp đỡ, chia sẻ vàđộng viên em trong suốt quá trình học tập tại trường Đại học Bách Khoa Hà Nội.

</div><span class="text_page_counter">Trang 12</span><div class="page_container" data-page="12">

<b> COPYRIGHT NOTICE</b>

I hereby that all the content presented in the report on the Nutrition Supply Systemfor ICU Patients is the result of the research and development process of our group.The data stated in the project is completely honest, reflecting the actual measurementresults. All information cited is subject to intellectual property regulations. Thereferences are clearly listed. We take full responsibility for the content written in thisproject.

Hanoi, Date … / 03 / 2023

<b>Student</b>

</div><span class="text_page_counter">Trang 13</span><div class="page_container" data-page="13">

<b>ABSTRACT </b>

The development of an automated deep learning-based software for reconstructing3D lung images and estimating Total Lung Volume (TLV) has the potential torevolutionize medical imaging and provide more accurate and efficient diagnoses forlung-related illnesses. In this study, the proposed software is curated mainly forVietnamese users with Vietnamese as the default language, allowing for greateraccessibility and ease of use. The system utilizes a modified Bi-directionalConvolutional (ConvLSTM) U-Net (BCDU-Net) neural network, which has proven tobe effective in medical image segmentation tasks.

To train and test the system, we utilized The Cancer Imaging Archive (TCIA)dataset, which is a large-scale dataset consisting of 9,593 thoracic CT scans of thelung. The dataset is divided into 36 training datasets, 12 live datasets, and 12 off-sitetest datasets. Our model is a modified extension of the current U-Net network, with theaddition of BCDU-Net and Densely Connected Convolutional layers. BatchNormalization was applied to increase the stability of the neural network and improveoverall performance.

The results of the study have shown that the proposed system achieves an accuracyof approximately 99% against actual data from commercial products, demonstratingthe potential of deep learning-based systems in medical imaging. Overall, the proposedsystem has the potential to improve medical diagnosis, especially in the field of lung-related illnesses, by providing more accurate and efficient analysis of 3D lung imagesand Total Lung Volume estimation.

</div><span class="text_page_counter">Trang 15</span><div class="page_container" data-page="15">

<b>3.1 Dataset...27</b>

<b>3.2 Pre-processing data...28</b>

<b>3.3 Deep Learning Based-model Segmentation...29</b>

<b>3.4 Calculator Lung Volume...31</b>

<b>3.5 3D Reconstruction...32</b>

<b>3.7 Experimental Results...34</b>

</div><span class="text_page_counter">Trang 16</span><div class="page_container" data-page="16">

associated with TLV obtained from PFTs. On the other hand, predictive equations havegrown into a prominent pathway to investigate TLV estimation from CT scans. Thishas been the area of interest in research for over a century, with the first relevant paperusing the gas dilution technique to demonstrate the correlation of externalmeasurements to the PFT. Previous papers were based on either the use of planimetrictechniques or the estimation of a specific geometry or several manual linearmeasurements. However, these approaches have two drawbacks which are relying onmanual measurements to estimate TLV and small sample sizes, resulting in unclearconclusions if the techniques could be generalized for different applications.

Following the above discussions and inspired by the CNN model, we study the roleof TLV labels based on thoracic CT imaging in deep learning training. We develop adeep learning-based software system by optimizing various state-of-the-art deep-learning approaches to automatically determine TLV using a large dataset of medicallung CT scans, curated for the Vietnamese market. Until now, there exists variouscommercial software worldwide [ CITATION Ste15 \l 1033 ] with similar workingmechanisms to aid medical specialists to calculate lung volume and simulate lungdiagrams. However, the cost has proven to be too high, and the language interfacemakes it difficult for local users who do not speak English. Moreover, such software isconstructed to be compatible with only the CT images produced by the company itself,and we could not attain access to the system’s actual accuracy and algorithm. To solvethis problem, we propose a localized software that applies a machine learningapproach for lung partitioning, 3D model reconstruction, and lung volume calculation.The effectiveness of our algorithm is evaluated by comparing calculated results fromthis research with actual patient data collected from Bach Mai hospital.

</div><span class="text_page_counter">Trang 17</span><div class="page_container" data-page="17">

<b> </b>

<b>CHAPTER 2. THEORETICAL THESIS</b>

In this chapter, the thesis provides an introduction to the knowledge and theoryutilized in the project, as well as the methods and models employed, including theirdefinitions and architectures. The first topic covered is image processing, which is thedirect focus of the problem that needs to be addressed. This is followed by artificialintelligence, which is the field utilized to tackle the problem, specifically through theU-net model using the deep learning technique of CNN.

<b>2.1 Image Processing </b>

<i><b>2.1.1 RGB Color System </b></i>

To begin with, a fundamental concept in image processing is the RGB color system.RGB represents the primary colors of light, which are Red, Green, and Blue, and thesecolors can be separated by a lens. By combining the three colors in different ratios, avast range of colors can be generated. This knowledge is crucial in image processingas it helps in understanding how colors are represented and manipulated in digitalimages.

<b> Fig 2.1 RGB color system</b>

The RGB color system is a fundamental concept in image processing, where RGBstands for Red, Green, Blue, the primary colors of light when separated from the lens.By combining these three colors in various proportions, different colors can beproduced. For each set of three integer values between 0 and 255, a unique color is

8

</div><span class="text_page_counter">Trang 18</span><div class="page_container" data-page="18">

generated. Since there are 256 possible values for each of the three colors, the totalnumber of colors that can be created in the RGB system is 16,777,216. This systemincludes popular colors such as:

(0, 0, 0) is black.(255, 255, 255) is white.(255, 0, 0) is red.(0, 255, 0) is green.(0, 0, 255) is blue.(255, 255, 0) is yellow.(0, 255, 255) is cyan.(255, 0, 255) is magenta.

Each variation in the values of red, green, and blue produces a unique color, but thedifference is usually so small that it's not noticeable to the naked eye. The definitiongiven in the previous paragraph describes the complete range of RGB. However, indigital video, the convention used for RGB is often not the entire range. Instead, videoRGB typically employs a scale of relative values, where a value of (16, 16, 16)represents black and (235, 235, 235) represents white.

<b>2.2 AI – Artificial Intelligence</b>

<b>2.2.1 Definition, history, and development of AI</b>

AI is a branch of computer science that strives to create intelligent machines thatcan perform tasks requiring human intelligence such as learning, perception, languagetranslation, and problem-solving. It has many applications in various fields such ashealthcare, finance, education, transportation, and more.

The history of AI dates back to the 1940s when scientists began studying machines'ability to perform calculations. The field of AI was established in the DartmouthConference in 1956, which focused on topics such as machine learning, naturallanguage processing, logic, and games. This conference marked the beginning of animportant research field aiming to create a computer program that could "think" like ahuman.

In the 1960s, AI algorithms and models were further developed, and John McCarthyproposed the concepts of "computer programmer" and "computer program," whichlater became the main terms in the field of AI. In the 1980s, artificial neural networkmodels were developed, making the field of AI even more powerful. Since then, AIhas become an important field in computer science and widely applied in many fields

</div><span class="text_page_counter">Trang 19</span><div class="page_container" data-page="19">

such as healthcare, finance, education, marketing, industry, and more. It is expectedthat AI will continue to play a significant role in solving human problems in the future.

<b>2.2.2 Deep Learning</b>

In the field of artificial intelligence, Machine Learning (ML) and Deep Learning(DL) are the two methods that are often mentioned first and are important methods forlearning from input data. ML allows computers to learn and improve from input datainstead of being programmed to perform a specific task. It can be said that machinelearning is teaching computers how to solve a specific problem by allowing them tolearn from data. This is done through the search for models, algorithms, and dataanalysis techniques, which help computers make predictions, pattern recognition, anddata classification. Common Machine Learning algorithms include supervisedlearning, unsupervised learning, and semi-supervised learning. Some applications ofMachine Learning include detecting credit card fraud, recommending products tocustomers, classifying spam emails, and facial recognition.

DL is a method of Machine Learning that uses artificial neural networks to learnfrom input data. Deep Learning uses multiple layers (or also called layers) of neuronsto learn and extract features from data. The deeper and more complex the featureextraction, the higher the quality of the output.

DL is applied in speech recognition, image and video recognition, natural languageprocessing, autonomous driving, and biotech robots.

<b>Fig 2.2 Relationship between AI, ML, and DL </b>

Although Deep Learning is considered a form of Machine Learning, these twomethods have fundamental differences such as:

10

</div><span class="text_page_counter">Trang 20</span><div class="page_container" data-page="20">

Complexity of input data: Machine Learning is usually used to process datawith medium complexity, while Deep Learning is often used for data with highcomplexity, such as images, videos, or natural language data.

Neural network architecture: Machine Learning uses traditional machinelearning algorithms such as SVM, Naive Bayes, Logistic Regression, K-NearestNeighbors (KNN), Decision Trees, and Random Forests. Meanwhile, DeepLearning uses a more complex neural network architecture, including multiplelayers of neurons, allowing it to learn complex features of data.

Amount of input data: To achieve high accuracy, Deep Learning requires moreinput data than Machine Learning. This helps Deep Learning learn morecomplex features, increasing the accuracy of predictions.

Processing speed: Deep Learning requires more computing power thanMachine Learning, so it requires more powerful processors and systemresources to process data.

<b>Fig 2.3 Comparision between Machine Learning and Deep Learning</b>

Deep learning uses artificial neural network algorithms to extract features and learn.Each algorithm has different complexities and is used for different types of problems.Some notable deep learning algorithms include:

Deep Neural Networks – DNN: Deep neural network

Convolutional Neural Networks – CNN: Deep learning convolutional networkAutoencoder – AE: Deep learning autoencoder

Recurrent Neural Networks – RNN: Deep learning network based on short-termmemory

Directional Deep Learning – DDL: Directional deep learning network

</div><span class="text_page_counter">Trang 21</span><div class="page_container" data-page="21">

Objective-Optimized Deep Learning – OODL: Deep learning networkoptimized for objectives

Self-adjusting Deep Learning – SADL: Deep learning network that self-adjusts

<b>2.3 CNN – Convolutional Neural Networks </b>

<i><b>2.3.1 Introduction to CNN </b></i>

Convolutional Neural Networks (CNN), also known as convolutional neural nets,are among the most advanced and popular Deep Learning models today, ideal forsolving problems with image data. It helps us build intelligent systems with highaccuracy like today. CNN is widely used in recognition, classification, andidentification of objects in images. The development of network architectures goeshand in hand with the development of computer hardware such as GPUs that are faster.Distributed and parallel training techniques on multiple GPUs allow a model to betrained in just a few hours, compared to training that used to take days and beexpensive. Deep learning supporting frameworks also appear more, improved andbecome tools to meet all necessary needs for deep learning training. The most popularcan be mentioned are three frameworks: PyTorch (Facebook), TensorFlow (Google),and MXNet (Intel) developed and backed by leading technology companies in theworld. Since the ImageNet dataset, image datasets have affirmed the role of driving thedevelopment of AI. Algorithms are compared to each other based on the leading resultsfrom standardized datasets. Thanks to the expansion of free training platforms likeGoogle Colab and Kaggle, everyone can access AI. The global strategy for AIdevelopment in corporations and countries around the world has led to the formationof AI research institutes that bring together many outstanding scientists andbreakthrough research.

<b> Fig 2.4 The development process of CNN [7]</b>

12

</div><span class="text_page_counter">Trang 22</span><div class="page_container" data-page="22">

Many new CNN architectures have been formed, developed and improved in termsof depth, block design, and block connectivity, such as GoogleNet, DenseNet, etc.Prior to 2012, most researchers believed that the most important part of a pipeline wasthe representation. SIFT, SURF, HOG were manually selected feature extractionmethods applied in combination with machine learning algorithms such as SVM, MLP,k-NN, Random Forest.

The characteristics of these architectures are:

The generated features cannot be trained because the rules that create them arefixed.

The pipeline is separated between feature extractors and classifiers.

A group of pioneering researchers believed that features can be learned through themodel, and to achieve complexity, features should be learned hierarchically throughmultiple layers. Because each image has features such as vertical, horizontal, diagonallines, etc., and even unique features that help identify objects. In the lowest layers ofthe CNN, the model has learned how to extract features similar to traditional featureextraction functions like HOG, SIFT. This research direction continues to developthrough the process of experimenting with new ideas, algorithms, and architectures. Atpresent, there are increasingly more CNN models being explored.

<b>2.3.2 CNN Architecture</b>

CNN, or Convolutional Neural Network, is a type of feedforward ANN that consistsof multiple layers of stacked convolutional layers. These layers use non-linearactivation functions such as ReLU and tanh to activate the weights in the nodes. Thelayers are connected to each other through convolutional mechanisms. The next layeris the result of the convolutional operation performed on the previous layer, thusallowing for local connections. Therefore, each neuron in the next layer is generatedfrom the result of a filter applied to a local region of the previous neuron. Each layeruses different filters, typically consisting of hundreds or thousands of filters, andcombines their results. There are also other layers such as pooling/subsampling layersused to filter out more useful information. During network training, the CNN modelautomatically learns values through filter layers. For example, in image classificationtasks, CNNs will try to find optimal parameters for the filters. The last layer is used toclassify the image. The main architecture of a CNN model consists of multiplecomponents connected together in a multilayer structure, including Convolution,Pooling, ReLU, and Fully connected.

</div><span class="text_page_counter">Trang 23</span><div class="page_container" data-page="23">

<b>2.3.2.1 Convolution layer </b>

The Convolution layer (Conv) is the most important layer in the architecture ofCNN. It is based on the theory of signal processing, and taking the convolution willhelp extract important information from the data. This operation can be visualized byshifting a window called a kernel over the input matrix, and the output value is thesum of the element-wise product of the kernel and the corresponding part of the inputmatrix. In image processing, the Conv operation can transform input information intocharacteristic features such as edges, directions, and color spots, and the kernel can beseen as a sliding window applied on the input image.

In CNN, there are several basic mechanisms that are commonly used, includingStride, Padding, and Feature map. Stride means shifting the filter map by a certainnumber of pixels from left to right. Padding is the adding of zeros to the input tomaintain the size of the feature map, and Feature map is the result of Convolutionoperation applied on the input. In processing, filters are shifted over the entire imagewith S (stride) steps, and in some cases, P pixels with a given color value (usually 0)are added around the edges of the image, which is called padding. Then, the outputfeature map is a matrix of size W2 x H2 x D2, where W, H, D represent the width,height, and depth of the output matrix.

The filter matrix of size (F x F x D1) + 1, where the additional 1 is the thresholdparameter of the filter, represents the weights, and these values are constant during theConv operation over the entire input image. This is an important property of the sharedweights that reduces the number of parameters that need to be learned during networktraining. As a result, the total number of parameters needed for processing can besignificantly reduced.

14

</div>