秉持尊重学生和家长隐私的原则，以下所展示的文字和图片，均已隐去涉及知识产权和敏感的信息。
还有N多学术素材，由于知识产权或家长原因，不便公开，敬请理解。
本文作者L PZ同学，品学兼优，目前是洛杉矶地区的中学男生，对计算机和数学非常有兴趣，经常参加洛杉矶当地的计算机和数学比赛。希望中学毕业后，能进入TOP30大学读书。为此，参与了美国名校科研项目，增加学术背景，开拓视野，获得真知。
本文是学生在美国大学科研学习时每天所写的感受。学生用稚嫩的笔画，勾勒出了自己的在大学参与科研学习的心路历程。学生经过与大学教授沟通，充分表达自己的研究兴趣后，经过协商选择了通过数学模型，利用计算机AI技术，分析解决现实生活中的医疗问题这一科研主题。从思路到文字阐述，很难想象一位中学生利用假期竟然能有如此高的科研成果。
由于学生参与的是与医学有关的科研主题，名校科研的老师在学生科研期间，额外安排了医学院教授的讲座，为学生补充、扩充生命科学类的基础知识。
在文章的结尾，附录学生在科研开始前的自我介绍PPT，和在报名选拔阶段参与美国大学面试后所写的感受，供其他学生及家长参考。
第一周
First day working at University, I get acquainted with the idea of AI SYSTEM. I did the self-introduction that day and basically no progress to show since it is my first day. I made a chart that illustrates the main difference between 2 different AI system.
System Aiming:
Technologies that change people's life
Improving people's life at the same time keep the balance between human and nature
The point which gave me a strong impression or interested me is
●Ranking systems of the surveillance in terms of the level of urgency
●Effective microorganism technology and environmental protection
Study US
1.smart mobility
Increase of electrical infrastructure construction
1212 electrical charging stations and 26498 electrical vehicles on the road
2.smart environment and energy
Street lights with adaptive control
SCOUP network: automatically tracks and reduces the greenhouse gas emission
Solar Panel
3.smart living
Net zero living
Use solar energy to provide all the electrical energy to vehicles and public areas
4.smart government
Open data portal
IoT platform to optimize the traffic condition
Study of Japan
1.The autonomous mobile dual arm robot(consumers’ demand)
2. Wind turbines
Mutual interactions between SYSTEM and real life
What benefits of SYSTEM can bring to our real life
Index differences between countries
I have been learning some basics about MatLab in order to get more familiar with using computer vision to do the image processing. Today I set the topic of my project which is the about cancer cell detection using computer vision. For the start of my project, I decided to work on familiarizing with background knowledge on cancer cells and current techniques that scientists have been using.
MATLAB learning progress:
ones(4) = 4*4 ones
zeros(4) = 4*4* zeros
diag(1 2 3 4 5)
reshape(1:10,[2,5]) reformat the sequence
NaN(2,4)
……
……
A day for the recent research update. I specify the class of cancer I would like to work on, which is esophageal cancer. I have been preparing the ppt which demonstrates the basic knowledge of cancer cells and how they grow.
Here are the sources that I used while creating the ppt:
https://www.google.com/search?q=how+cancer+cells+form+and+spread&rlz=1C1CHBF_enUS746US746&source=lnms&tbm=isch&sa=X&ved=0ahUKEwjir7af0tPbAhUEbq0KHfCtCEsQ_AUICygC&biw=1536&bih=734#imgrc=2XP2jZZneqjCMM:
……
……
Esophageal cancer
The choice of lenses for the endoscopic devices:
CCD(broadcast) and CMOS(phone) lenses
CCD can present high definition images and has a high rate of updating, but has the problem of blooming------it uses lots of energy and holds too many pixels at once
CMOS presents low definition images and has low rate of updating, but has no problem of blooming------it captures each pixel individually
This is an important link(an article) that illustrates how did they use computer vision on esophageal cancer detection:
……
Annotations on the paper will be put in my dropbox
#meaningful region → suspicious tissues
Link that illustrates the steps of processing cell images:
……
Today I focused on the learning of neural network since I do not know how it works. I learned it through watching videos on the YouTube and people who attended the meeting also talked about the neural network and CNN.
28 x 28 -> 784 pixels, 784 neurons
Weight = brightness of neurons
Bias = showing of edge
Multilayer perception (old technology)
According to the cost, it needs adjustment on the weights and biases. After adjusting the indexes, the computer learns faster than before
After learning about neural network, I know that I need a huge database to train the network. However, I need to try the algorithm with an image first. I searched the algorithm online that can detect cells. There is a kind of technique called K-means clustering, which can extend out the features clearly in cells’ images. The codes are following below:
lab_I = rgb2lab(I);
……
Clustering is a way to separate groups of objects. K-means clustering treats each object as having a location in space. It finds partitions such that objects within each cluster are as close to each other as possible, and as far from objects in other clusters as possible. K-means clustering requires that you specify the number of clusters to be partitioned and a distance metric to quantify how close two objects are to each other.
Different layers analyze different specific features
……
I prepared a presentation that introduces some knowledge about cancer cells. My ppt contains both the background knowledge and image processing technique. I used K-means clustering to clearly show the blue nuclei. For the following work, I will work on different cases of cancer cells and see if the algorithm is still working. Also I will consider if the live detection is possible or not.
https://www.google.com/search?q=esophageal+cancer&rlz=1C1CHBF_enUS746US746&oq=esophageal+cancer&aqs=chrome.0.69i59l2j0l4.4429j0j7&sourceid=chrome&ie=UTF-8
……
Why choose K-mean method:
For the segmentation of background cells, k-means segmentation algorithm is used because it performs better in comparison to other commonly used segmentation methods. In feature extraction phase, it is proposed to extract various biologically interpretable and clinically significant shapes as well as morphology based features from the segmented images.
……
Improvements on the presentation:
Put more statistics that can prove that cancer cell detection using CV is more reliable and efficient than old method of detecting
第二周：
In order to extend out the features, we need to separate the nuclei from the background.
Healthy cells
I analyzed the color of nuclei section. Then I printed the RGB range of nuclei color.
Any area’s RGB values that are not in the range of nuclei RGB range will be shown black.
Then I use canny edge detection to make the edge more obvious.
The result is like this:
Comparison
Professor H's lecture:
Questions for the professor
1.How to differentiate cancer cells and healthy cells based on their microscopic images?
Cancer cells’ nuclei are larger
Shape of the cells
Cancer cells are disorganized
2.Do different tissues have different shapes of cancer cells?
Yes
3.Do different tissues have different layout or distribution of nuclei?
Yes
4.Do healthy tissues have partial mutations?
Maybe not, because the body is constantly getting rid of cancer cells. However, the mutations are always happening in our cells.
5.Are pathologists 100% true about the diagnosis on the cancer cells?
Yes by histography but not by H&E images
6.What kind of stain do they use after the biopsy?
H&E and histography
7.What is the current technology and situation?
Incomplete to use AI and CV
6/23/18 lecture notes
Esophagus
Oncogene stimulates cell growth, so they should be eliminated.
Cancer cells cure(killing specific cells):
1.insert suicide cells and kill the disease cells
2.insert the marker cells and attracts the immune cells
chemotherapy: some chemicals inhibit the disease cells, but the effect is very universal. That is why there so many side effects. (not supported)
Special Case: take out the tumor genes and insert the suicide cells into the tumor cells. Then put the cells back to the tumor tissue.
research update
The next step:
1.Test the current feature detection
2.detect the cytoplasm(show the edge of the cells)
Features that can prove the cancer cells
1.the shape of nuclei
2.the distribution of nuclei
3.the number of nuclei in a cell
H&E stain:
Hematoxylin and eosin stain
#the most widely used stain in medical diagnosis
#the combination of chemicals produces blues, purples and reds
#DNA/ RNA (nuclei) are dyed violet.
#Nuclei are in purple
#Cytoplasm is in pink
Comparison
After canny detection:
Healthy cell
Esophageal cancer cell
Esophageal cell with dysplasia and without dysplasia
PPT Links:
https://www.verywellhealth.com/cancer-cells-vs-normal-cells-2248794
……
第3、4周
I prepared some detailed background knowledge about the esophageal cancer. This summary basically covers all the knowledge I think people need to know about esophageal cancer and its characteristics.
Esophageal Cancer
1. About esophagus
The esophagus is a 10-inch long, hollow, muscular tube that connects the throat to the stomach. It is part of a person’s gastrointestinal (GI) tract. When a person swallows, the walls of the esophagus squeeze together to push food down into the stomach.
What is GI tract?
The GI tract is a series of hollow organs joined in a long, twisting tube from the mouth to the anus. The hollow organs that make up the GI tract are the mouth, esophagus, stomach, small intestine, large intestine, and anus.
2. About the esophageal cancer
Cancer begins when healthy cells change and grow out of control, forming a mass called tumor. A tumor can be malignant (恶性) or benign (良性). A cancerous(malignant) tumor can spread it to other organs in the body. A benign tumor can grow but cannot spread.
Esophageal cancer begins in the inner layer of the esophagus and gradually spreads out to lymph nodes (tiny bean-shaped organ that fights infection), as well as to the blood vessels in the chest.
3. Types of esophageal cancer
……
Glands: organs that secrete
4. Statistics
In 2016, estimated 17,290 adults in the US will be diagnosed with esophageal cancer. The disease accounts for 1% of cancer diagnosed in the US. It is diagnosed more often than other parts of the world. It was estimated that 15,850 deaths will occur from this disease this year. Esophageal cancer is the seventh most common cause of cancer death among men.
5-year survival rate tells you how many people out of 100 people will survive in 5 years after the cancer is found. The 5-year survival rate of esophageal cancer is 19%
The 5-year survival rate of people with cancer located only in the esophagus is 43%. The 5-year survival rate for those with disease that has spread to surrounding tissues or organs and/or the regional lymph nodes is 23%. If it has spread to distant parts of the body, the survival rate is 5%.
5. Medical illustration
GERD
#Diet: A diet that is low in fruits and vegetables and certain vitamins and minerals can increase a person's risk of developing esophageal cancer
#Being severely overweight and having too much body fat can increase a person's risk of developing esophageal adenocarcinoma
7. Screening
The goals of screening are to:
#Lower the number of people who die from the disease, or eliminate deaths from cancer altogether
#Lower the number of people who develop the disease
People with Barrett's esophagus may be advised to have regular endoscopic examinations. An endoscopic examination is a procedure that used a flexible, lighted tube to look inside the esophagus. During these examinations, biopsies can be taken to remove a small amount of tissue for examination under a microscope.
8. Signs and Symptoms
People with esophageal cancer may experience the following symptoms or signs. Sometimes, people with esophageal cancer do not have any of these changes. Or, the cause of a symptom may be another medical condition that is not cancer.
……
9. Diagnosis
Doctors use many tests to find, or diagnose, cancer. They also do tests to learn if cancer has spread to another part of the body from where it started. If this happens, it is called metastasis.
This list describes options for diagnosing this type of cancer, and not all tests listed will be used for every person. Your doctor may consider these factors when choosing a diagnostic test:
……
Biopsy
A biopsy is the removal of a small amount of tissue from the suspicious area for examination.
10. Stages
TNM staging:
# Tumor(T): How deeply has the primary tumor grown into the wall of the esophagus and the surrounding tissue?
#Nodes(N): Has the tumor spread to the lymph nodes? If so, where and how many?
#Metastasis(M): Has the cancer metastasized to other parts of the body? If so, where and how much?
The purpose of staging is to make a common standard for doctors so that they can better diagnose patients.
Tumor(T):
I think this week is the most struggling week for me since it is so hard to solve the professional and technical problems. It is still very difficult for me to tell the direct difference between a cancerous cell and a normal cell. The result of image processing is also not very clear. The result is kind of far from the what I have expected. I have been trying to extend out the nuclei in the images, but it only worked for a few special images. Therefore, I decided to build up a neural network and directly import my data into it without any further processing (knew that it will not have a very good result).
I access to the database of National Cancer Institute
https://portal.gdc.cancer.gov/
Two types of code that can do the image processing:
PPT links:
https://www.google.com/search?rlz=1C1CHBF_enUS746US746&biw=1536&bih=734&tbm=isch&sa=1&ei=aK07W_KaFeGD5wK27qTQAQ&q=G1+esophageal+cancer+in+H%26E+image&oq=G1+esophageal+cancer+in+H%26E+image&gs_l=img.3...2257.6402.0.6829.17.15.2.0.0.0.724.1598.10j1j6-1.12.0....0...1c.1.64.img..3.0.0....0.bYDEYoH9x2U#imgrc=3Rv3dsqrGhj73M:
Here are six methods I summarized that can basically accomplish the goal of image processing.
Method 3, 4 and 6 are the ones I decided to be trained in the neural network.
method 1:
1. purple color exclusion
2. binarization
3. canny edge detection
……
method 6:
1. binarization
2. erosion
3. canny edge detection
The choice of neural network is the most important part of machine learning.
Kohonen Self Organizing Neural Network
This self organization process has different parts, in the first phase every neuron value is initialized with a small weight and the input vector. In the second phase, the neuron closest to the point is the ‘winning neuron’ and the neurons connected to the winning neuron will also move toward the points
Kohonen Neural Network is used to recognize patterns in the data. Its application can be found in medical analysis to cluster data into different categories.
Convolutional Neural Network
In this neural network, the input features are taken in batch wise like a filter. This will help the network to remember the images in parts and can compute the operations. These computations involve conversion of the image from RGB or HSI scale to Gray-scale. Once we have this, the changes in the pixel value will help detecting the edges and images can be classified into different categories.
学生在报名选拔阶段，面试后所写感受：
Dear mentor,
I finished the interview on Saturday morning. I think it goes well in general, but there are some terms and understands that are far beyond my knowledge. I need to make lots of improvements. The interviewer first asked me about whether I am good at linear maths and how is my math level. I said I am taking advanced math courses right now, but the interviewer wanted to know more about my background experience on applied mathematics. For the experience, I am in one of our Vex robotics team right now and my team has qualified for states competition. I told him that the only part that uses linear mathematics and data references on our robot is the sensor. I briefly illustrated how our sensors work and what are their functions. Other than that, the design and construction of the robot need some of the knowledge of physics. My interviewer then highly suggests me do some research and attend the International Mathematical Contest In Modeling. He told me that this competition is based on the applied mathematics. After this question, he asked me about my programming background experience. I said that I am in robotics team, so I knew little about Vex robotics coding which is C language based. I also told him that I am taking AP computer science principle right now and we have done projects based on language Python. At the start of the semester, I also did a web page project using language HTML. The web page is about introducing myself and creating links inside my introduction. Then he told that python is really popular language right now, but I need to work on the learning of the Java. Java is also the most useful programming language right now. After this question, he asked me if I have heard of this term “computer vision”. At first, I was not sensitive about this term. He suggested me search this term and he gave an example of computer vision which is AI.
After providing the example of AI, I said that I have heard of, but not fully understand what is means. He asked me whether I can give some examples of computer vision. My answer was self-driving cars, because it is one of the most heated technological topics in our modern world. And self-driving cars are also similar to AI, which is about computers gain understanding based on the digital images they capture. Then he asked if I have read any academic papers about computer science. I said that I have not. He highly suggested me take a look at the two web pages. The first one is called Web of Science, and the other one is called Engineering village. There are lots of academic papers on both of the websites and I can learn some from those papers. At the end of the interview, I asked what kind of preparation should I do if I am in this program. My interviewer is very nice and eager to teach me. He gave a lot of useful advices. Although I may not very full competent for this program, I would like to meet the challenge and learn more about computer science and coding.
学生在抵达美国大学前，做的自我介绍文件
北京博师屯儿教育科技有限公司（博士屯教育），注册在北京，团队管理者平均行业经验8年。专注美国学界合作，目前合作大学遍布全美，深度合作区域有波士顿区域，纽约区域，旧金山区域等。
博士屯教育，专注美国名校合作，为学生定制有利于成长的一站式背景提升方案，助力本硕博名校录取。
博士屯，就是做背景提升的，目标就是提高本、硕、博录取质量。
As the most energetic provider of research educational services in China and North America, Ph.D village offers education for helping students to get into better college and helping improve their quality of life. Our wide range of educational programs, services and products includes research program, elite program, visiting scholar program, as well as paper online education.
Ph.D village has become one of the most recognized brand in research education area.
