VIETNAM NATIONAL UNIVERSITY, HANOI INTERNATIONAL SCHOOL GRADUATION PROJECT USING AI AND RAMAN SPECTROSCOPY TO MEASURE GLUCOSE Student’s name Dang Phuong Thao Hanoi - Year 2024 VIETNAM NATIONAL UNIVERSITY, HANOI INTERNATIONAL SCHOOL GRADUATION PROJECT USING AI AND RAMAN SPECTROSCOPY TO MEASURE GLUCOSE SUPERVISOR: PhD. Nguyen Thanh Tung STUDENT: Dang Phuong Thao STUDENT ID: 19071628 COHORT: QHQ2019 SUBJECT CODE: INS401401 MAJOR: Informatics and Computer Engineering Hanoi - Year 2024 2 THANKS Associate Professor Dr. Nguyen Thanh Tung has my deepest gratitude for his constant commitment, direction, and encouragement during my capstone project. His knowledge, support, and enlightening criticism were invaluable in determining the course and outcome of my project.
I also want to express my sincere gratitude to each and every one of the distinguished instructors at International University - Vietnam National University. Their unceasing attempts to transfer theoretical and practical knowledge have been extremely helpful to my development on both a personal and professional level. Their dedication to quality has given me a strong foundation of knowledge and abilities that will help me in all of my future undertakings. When I look back on my path, I realize that this project, which is the pinnacle of my academic endeavors, marks an important turning point.
It has given me the chance to test and improve my skills in addition to showcasing the knowledge I have gained. Through diligent study and application of teachings from various sources, including lectures, textbooks, and scholarly works, I have consolidated my learning and expanded my horizons. This project has been a test of academic proficiency and a journey of self- discovery and personal growth. I am profoundly grateful for the support and guidance that have enabled me to navigate this transformative experience with confidence and determination.
3 ABTRACT This research study investigates the utilization of Convolutional Neural Networks (CNN) for the purpose of categorizing and comparing outcomes among samples in order to create an artificial intelligence system for diabetes assessment. The dataset consists of data obtained from authentic persons in Vietnam using a Raman spectrometer, after eliminating background noise. The study developed a system that attained a perfect accuracy rate of 95% in categorizing data with two labels, and an approximate accuracy rate of 84.4% in categorizing data with three labels. The system effectively distinguished between samples that were positive and negative for diabetes.
The project entailed methodical alterations to different system elements, including the quantity of training courses, samples per training session, and the configuration of each training class. The best setup exhibited improved accuracy for the provided dataset. The purpose of these findings is to enhance the field of diabetes screening and early detection. Tackling such a crucial subject requires a tremendous commitment and allocation of resources, including significant investments in time, staff, and equipment.
The extensive range of this project includes many stages, such as sample acquisition, sample analysis, development and training of machine learning models, and design of testing devices, among other tasks. Nevertheless, this project has a unique emphasis on constructing and instructing the machine learning model, avoiding elements such as measurement and data acquisition, device design, and procedural follow-up procedures. 4 Table of Contents THANKS .1 What is diabetes? .2 Types of Diabetes .2 AI in the Medical Field .4 Convolutional Neural Networks (CNN) .7 Accuracy, Loss, Validation Accuracy, and Validation Loss .1 What is Raman Spectroscopy? .2 History of Raman spectroscopy .3 Theory of Raman Spectroscopy .4 Applications of Raman Spectroscopy in healthcare .1 Breast cancer detection and diagnosis .2 Analytical Quality Control in a Hospital Environment .1 Type of Model .1 Shuffle data function .2 Plot history function .3 Save graph function .2 Three-labels Classification .2 Three-labels Classification .59 6 Table of Figures and Tables Figure 1: Prevalence of diabetes between 2010-2030. 8 Figure 2: Type of Diabetes.
10 Figure 3: Structure of CNN. 17 Figure 4 Supervised Learning. 20 Figure 5 Chandrasekhara Venkata Raman 1888-1970. 25 Figure 6 Import library.
36 Figure 7: Shuffle data function. 38 Figure 8: Plot figure. 40 Figure 9 Save graph. 41 Figure 10 Print data to terminal.
42 Figure 11 Prepare data. 42 Figure 12 Data after shuffle. 43 Figure 13 Print other information. 44 Figure 14 Using TensorBoard tool.
44 Figure 15 Loss function for two labels. 45 Figure 16 Design model layer for two labels. 46 Figure 17 Use Flatten layer, Dropout layer to prevent overfitting. 47 Figure 18 Using sigmoid activation.
48 Figure 19 Design model layer for three labels. 49 Figure 20 Using Softmax activation. 50 Figure 21: 2-data labels result. 51 Figure 22: Plot Accuracy and Loss for 2-data labels.
52 Figure 23: 3-data labels result. 53 Figure 24: Plot Accuracy and Loss for 3-data labels. 54 Table 1 Meaning of binary Classification. 34 Table 2 Meaning of three-labels Classification.
35 Table 3 2-labels training data results. 52 Table 4 3-labels training data results. 59 7 Chater 1: Overview Diabetes mellitus, commonly known as diabetes, is a metabolic disorder in which the body cannot tolerate glucose, leading to higher blood sugar levels than normal. Glucose is vital for maintaining the body's well-being as it serves as the primary fuel required for the proper operation of cells, particularly those in the brain.
The etiology of diabetes is multifaceted, contingent upon the particular kind of diabetes. However, irrespective of the kind, the condition causes elevated amounts of glucose in the blood, leading to a variety of serious health complications. In the past, type 2 diabetes primarily impacted adults, whereas type 1 diabetes was more prevalent among young individuals. Currently, there is a growing prevalence of type 2 diabetes among youngsters.
Furthermore, it is common for young patients to have problems soon after the commencement of the disease, such as vascular diseases. Certain situations are disregarded or ignored because the symptoms are either absent or not severe. Prompt identification and timely intervention are crucial for individuals with diabetes. By effectively managing type 1 diabetes, one can prevent the development of type 2 diabetes, hence reducing the potential difficulties associated with the condition.
Figure 1: Prevalence of diabetes between 2010-2030 8 Presently, the number of individuals diagnosed with diabetes in Vietnam stands at over seven million. Significantly, more than 55% of patients encounter difficulties, including 34% who suffer from cardiovascular complications, 39.5% who experience eye and nerve complications, and 24% who face kidney complications. Complications associated with diabetes not only escalate healthcare expenditures but also diminish the quality of life. As per the World Health Organization (WHO), some 422 million individuals are affected by this illness, with the majority living in countries with low to middle incomes.
The annual mortality rate also surpasses 1.5 million individuals, a worrisome statistic. Over the past few decades, there has been a consistent and gradual rise in both the incidence and mortality rates. Invasive testing, particularly blood testing, is the prevailing diagnostic procedure employed nowadays. Despite its high level of accuracy, this approach is constrained by its cost and the time it takes to get findings, which might cause discomfort for patients.
Consequently, non-invasive testing approaches are attracting considerable interest. We are engaged in a collaborative effort with Associate Professor Dr. Nguyen Thanh Tung to investigate and develop a specific methodology. We employ Raman spectroscopy and artificial intelligence (AI) to deliver prompt outcomes, surpassing the constraints of conventional approaches.1 What is diabetes? Diabetes is a persistent medical illness that occurs when the pancreas does not create enough insulin, or when the body is unable to properly use the insulin, it produces.
Insulin is a hormone responsible for the regulation of glucose levels in the bloodstream. Diabetes causes an inability of the body to regulate blood sugar levels, resulting in high blood sugar as a result. Over a period of time, this can cause significant harm to several biological systems, particularly the nerves and blood vessels.2 Types of Diabetes Diabetes can be classified into three primary categories: type 1, type 2, and gestational diabetes. The majority of patients receive a diagnosis of either type 1 or type 2.
Figure 2: Type of Diabetes 2.1 Type 1 Diabetes This is the onset phase of a patient's diabetes journey, characterized by the inability of the body to synthesize insulin (the hormone responsible for regulating blood glucose levels). On average, approximately 5-10% of people are diagnosed with type 1 diabetes. Due to improvements in the food business, the diagnosis of type 1 diabetes can now occur at different ages and its onset might be sudden. These symptoms are typically 10 subtle, making them challenging to identify unless examined on a regular basis.
This is also the underlying factor responsible for the development of type 2 diabetes.2 Type 2 Diabetes The majority of patients currently diagnosed with diabetes have type 2 diabetes, accounting for approximately 90-95% of cases. In this stage, the body is unable to effectively utilize insulin to regulate blood sugar levels, necessitating medical intervention such as medications and injections. Type 2 diabetes develops gradually over several years and typically affects adults, although there has been a rise in cases among younger individuals due to excessive consumption of sugar through processed foods. Similar to type 1 diabetes, type 2 diabetes lacks specific symptoms and requires regular screening.
However, adopting healthier dietary choices and engaging in physical exercise can help delay the onset of type 2 diabetes.3 Gestational Diabetes Gestational diabetes occurs in pregnant women who do not have pre-existing diabetes. This issue commonly arises during pregnancy and typically resolves after giving delivery. Gestational diabetes occurs when the placenta of the mother generates hormones during pregnancy that result in the accumulation of glucose in the blood. This is accompanied by inadequate synthesis of insulin, which is necessary to regulate sugar levels.
While this disease may resolve itself after giving birth, it can still adversely affect the child's health and heighten the likelihood of acquiring type 2 diabetes.1 Definition Artificial intellect (AI) is a computer science discipline that focuses on addressing cognitive challenges often associated with human intellect. Artificial Intelligence possesses the capacity to replicate human-like behavior in domains such as cognition, innovation, and visual perception. The objective of artificial intelligence (AI) is to develop autonomous systems that possess the ability to comprehend the significance of information and subsequently utilize the acquired knowledge to address issues in a way 11 akin to human beings. Artificial intelligence (AI) constantly acquires new knowledge by leveraging accurate incoming data.
ChatGPT is an artificial intelligence software that is continuously improving its ability to respond to extremely intricate questions. This is the outcome of its ongoing process of acquiring knowledge. Artificial Intelligence (AI) can be utilized across a wide range of domains, including both everyday life and industrial settings, to streamline corporate operations, improve consumer satisfaction, and foster groundbreaking advancements.2 AI in the Medical Field Artificial Intelligence (AI) is predicted to possess transformative capabilities in the medical domain and has achieved notable advancements in the treatment of diseases. AI, equipped with sophisticated algorithms and rapid data analysis capabilities, can aid in early-stage disease identification, resulting in expedited treatment and improved patient outcomes.
Due to the substantial advantages, numerous studies have been conducted to utilize artificial intelligence in the field of diagnosis and therapy. A recent study conducted by a team of experts from the University of Canterbury in New Zealand has demonstrated that the utilization of artificial intelligence (AI) can assist healthcare practitioners in devising more efficient tactics for cancer treatment, ultimately enhancing the likelihood of saving patients' lives. Associate Professor Alex Gavryushkin of the Mathematical Biology Research Centre of the University of Canterbury performed a four-year investigation, and this is the conclusion. The experts devised algorithms to scrutinize biological data pertaining to intricate genetic disorders, such as cancer and gout, with the aim of formulating therapy procedures grounded in genetic information.