Computing Research Project - AI, Thesis of Introduction to Computing

Download Computing Research Project - AI and more Thesis Introduction to Computing in PDF only on Docsity!

ASSIGNMENT 1 FRONT SHEET

Qualification BTEC Level 5 HND Diploma in Computing

Unit number and title Unit 13 :Computing Research Project

Submission date November 05th,2019 Date Received 1st submission

Re-submission Date Date Received 2nd submission

Student Name Duong Manh Quynh Student ID GCD

Class GCD0601 Assessor name Phan Thanh Tra

Student declaration

I certify that the assignment submission is entirely my own work and I fully understand the consequences of plagiarism. I understand that

making a false declaration is a form of malpractice.

Student’s signature Duong Manh Quynh

Grading grid

P1 P2 P3 P4 P5 M1 M2 M3 D1 D

Subject: Computing Research Project

 Summative Feedback:  Resubmission Feedback:

Grade: Assessor Signature: Date:

Internal Verifier’s Comments:

Subject: Computing Research Project

Subject: Computing Research Project

Artificial intelligence in reproductive medicine- Selection embryos and oocytes

Author: Duong Manh Quynh

Subject: Computing Research Project

• Literature Review Contents
• Research Proposal Form
  • Proposed Title:.......................................................................................................................................
• Artificial Intelligences in reproductive medicine- selection embryos system.
  • Introduction about system
  • Research methodologies
    • Primary and secondary research......................................................................................................
    • Quantitative and qualitative research..............................................................................................
    • Chosen the appropriate for the research
  • Design study
  • Artificial Intelligence system
    • Segmentation and preprocessing
    • Feature extraction..............................................................................................................................
    • Classification
    • Statistical analysis
  • RESULT
  • DISCUSSION
• CONCLUSION
• 1. Bibliography

Subject: Computing Research Project

Literature Review According to (Saeedi P, Yee D, Au J, Havelock J., 2017), accurate evaluation of embryo survival is a main component in maximizing pregnancy rates and optimizing IVF treatments. In all most of case, the embryologists will select embryos or oocytes by noninvasive examination based on visual observations focus on morphology and dynamic development during blastocyst stage. Because, the evaluation of embryos is subjective, so subject to change between and internal observers considering the existence of the embryo scoring system and the embryologist’s experience and expertise on the final success. (Manna C1, Nanni L, Lumini A, Pappalardo S., 2012). Besides, the potential consequences of multiple pregnancies and the risk of serious complications include pre-eclampsia, maternal hemorrhage are increased as to transfer the work increased (Bromer JG1, Seli E., 2008). To automatic the embryo selection, (Santos Filho E1, Noble JA, Poli M, Griffiths T, Emerson G, Wells D., 2012) proposed a method or segmenting images and classifying images of human blastocysts with semiautomatic classification. They trained two SVM classifiers to classify internal cell mass (ICM) and trophectoderm (TE) quality. By calculating the fractal dimensions and the average thickness of the TE and ICM image texture descriptions, the main morphological characteristics of blastocysts were clearly characterized. In addition, adjusting a microscope such as greater contrast and stronger boundaries of individual features can yield better image analysis. (Singh A, 2015) presented a new algorithm in a completely automated method to identify TE regions of human blastocysts. They used the Retinex algorithm to improve the quality of the input image, eventually achieving an average shape accuracy of 87.8% to detect TE regions. (Saeedi P, Yee D, Au J, Havelock J., 2017) introduced the first automated method to divide TE and ICM together in human blastocyst images. Creating and testing data sets of 211 blastocyst images at different levels, they reported an accuracy of 86.6% for TE determination and 91.3% for ICM. Embryo morphology is still the current tool to select embryos for transfer. Data obtained from automatic image recognition can provide the opportunity to objectively evaluate embryos and analyze them in a more quantitative manner. Methods of selecting embryos based on published morphological parameters. Moreover, embryo assessment using a dynamic monitoring system (Time-Lapse (TL)) provides continuous information about embryo developmental and morphological stages, although timeless algorithms are still in doubt. (Storr A1, 2018) Some researchers do not see this as evidence of the benefits of embryo selection (Kaser DJ1, Racowsky C2., 2014) (Armstrong S1, 2018). (Carrasco B1, Arroyo G2, Gil Y2, Gómez MJ2, Rodríguez I2, Barri PN2, Veiga A2,3, Boada M2., 2017)retrospective analysis of 800 human embryos with known implant data in an incubator with a Time-Lapse system. They developed a model based on morphological analysis and evaluation of embryos morphology on D3. The morphokinetic can eliminate embryos with the lowest implantation potential. Based on the overview about the selection embryos, this studies will focus on the work that concentrates the efforts on the possible prediction of the quality of the embryos and oocytes in order to improve the performance of assisted reproductive technology, starting from their images. I have many reason to choose this title. One of the important reason is my general. As you know, IVF (In Vitro Fertilization) is one of the fastest growing medical fields. The first birth after successful treatments was reported in 1978 (Steptoe PC, Edwards RG., 1978). In addition, according a report in 2014, in 2005, more than 1 million cycles of treatment were performed worldwide and in that year alone, more than 250,000 babies were born with in vitro fertilization (Zegers-Hochschild F1, Mansour R2, Ishihara O3, Adamson GD4, de Mouzon J5, Nygren KG6, Sullivan EA7., 2014). Although there are still issues related to treatments that need to be addressed, many aspects of in vitro fertilization have improved significantly, the relatively low implantation rates often result in a number of embryos and lead to multiple pregnancies.

Subject: Computing Research Project

Research Proposal Form Student name Duong Manh Quynh^ Student number: GCD Centre name: University of Greenwich Date: October 27th, 2019 Tutor: Phan Thanh Tra Unit: Proposed Title: Artificial Intelligence in Reproductive Medicine- Embryo and oocytes selection Section One: Title, objective, responsibilities Research question: How Artificial Intelligence based systems help IVF in selection embryo and oocytes? Objectives: I want to learn:  What is the AI techniques that applying in reproductive medical- in which selection embryo and oocytes?  How AI work in selection embryo and oocytes?  Section Two: Reasons for choosing this research project Reason for choosing the project:  I am interesting in the work of AI in medical area, special reproductive medical.  Find a new way in the IVF for the woman who want to become a real mother  Prevent the woman before the risks that IVF bring Section Three: Literature sources searched  Timo Ojala, Matti Pietikäinen, Topi Mäenpää, 2002. Multiresolution Gray-Scale and Rotation Invariant Texture Classification with Local Binary Patterns. [Online] Available at: https://www.computer.org/csdl/journal/tp/2002/07/i0971/13rRUEgarCe  Zegers-Hochschild F1, Mansour R2, Ishihara O3, Adamson GD4, de Mouzon J5, Nygren KG6, Sullivan EA7., 2014. International Committee for Monitoring Assisted Reproductive Technology: world report on assisted reproductive technology, 2005.. [Online] Available at: https://www.ncbi.nlm.nih.gov/pubmed/

Subject: Computing Research Project

Artificial Intelligences in reproductive medicine- selection embryos system. Introduction about system To select the highest quality embryos for transfers based on morphological analysis is the methods that used frequently. Many morphological scoring systems have been proposed and considered for selections appropriate embryos for transfers (Puissant F1, Van Rysselberge M, Barlow P, Deweze J, Leroy F., 1987). To select the highest appropriative embryos for transfer can achieve by human embryos culture extending to blastocyst stage (David K, Gardner, D.Phil.,$ Pauline Vella, B.Sc., Michelle Lane, Ph.D.,*Lyla Wagley, M.Sc.,$ Terry Schlenker, M.A., $ and William 6. Schoolcraft, M. D.$-, 1997).  C Manna1, G Patrizi2,4, A Rahman3, H Sallam3, 2004. Experimental results on the recognition of. [Online] Available at: https://www.rbmojournal.com/article/S1472-6483(10)60931-5/pdf Section Four: Activities and timescales

1. Collect materials concentrate with the topic research.
2. Complete research proposal
3. Milestone 1 [28-10] : Get feedback from Tutor about the research proposal
4. Produce a project plan:  WBS  Gantt Chart
5. Writing a literature review and some part of report  AI in reproductive medical- selection embryos and oocytes  Research method  Chosen research method
6. Check project progress: research proposal, plan, literature review
7. Milestone 1 [30-10] : Get feedback from Tutor about the primary research.
8. Conduct primary research
9. Milestone 1 [01-10] : Get feedback from Tutor about the result
10. Make experiment and collection some report
11. Writing assignment 1 which contains LO1, LO
12. Milestone 1 [05-10] : Submit assignment 1 Section Five: Research approach and methodologies Research approaches: Scrum Research methodologies: Record keeping and Experiment

Subject: Computing Research Project

Figure 1 Primary and secondary research. (McCrocklin, 2018) Primary research Primary research fills the next gaps of information that a researcher cannot collect through secondary research methods. The objective of the study is to answer specific questions directly related to the project. This type of research is extremely valuable, however, due to its nature, takes longer to collect than secondary research. Example for primary research:  Interview  Online survey  Focus group  Observations Secondary research Secondary research includes research papers published in research reports and similar documents. These documents may be provided by public libraries, websites, data obtained from surveys, etc.Some governmental and non-governmental agencies also store data, which can be used by the public. For research purposes and can be accessed from them. Secondary research is much more effective than the main study, because it uses existing data, unlike the main study where data is collected by organizations or businesses or they can hire third parties. to collect data on their behalf. Example for secondary research:  Data available on the internet

Subject: Computing Research Project

 Government and non-government agencies  Public libraries  Education institutions  Commercial information sources Key differences between primary and secondary research Primary Research Secondary Research Research is conducted first hand to obtain data. Researcher “owns” the data collected. Research is based on data collected from previous researches. Primary research is based on raw data. Secondary research is based on tried and tested data which is previously analyzed and filtered. The data collected fits the needs of a researcher, it is customized. Data is collected based on the absolute needs of organizations or businesses. Data may or may not be according to the requirement of a researcher. Researcher is deeply involved in research to collect data in primary research. As opposed to primary research, secondary research is fast and easy. It aims at gaining a broader understanding of subject matter. Primary research is an expensive process and consumes a lot of time to collect and analyze data. Secondary research is a quick process as data is already available. Researcher should know where to explore to get most appropriate data. Table 2 Key differences between primary and secondary research. (Bhat, 2018)

Quantitative and qualitative research

Qualitative research Qualitative research is mainly exploratory. It is used to gain an understanding of basic reasons, opinions and motivation. It provides insight into the problem or helps develop ideas or hypotheses for potential quantitative research. Qualitative research is also used to explore trends in thoughts and ideas, and to drill deeper into the issue. Methods for collecting qualitative data vary using unstructured or semi-structured techniques. Some common methods include focus groups (focus group discussions), personal interviews and participation / observations. The sample size is usually small and respondents are chosen to complete a certain quota. Quantitative research Quantitative research is used to quantify problems by creating numeric data or data that can be converted into usable statistics. It is used to quantify attitudes, opinions, behaviors and other defined variables - and to generalize results from a larger sample population. Quantitative research uses measurable data to form events and discovery models in research. Quantitative data collection methods are much more structured than qualitative data collection methods. Quantitative data collection methods include many different forms of surveys - online surveys, paper surveys, mobile surveys and kiosk surveys, face-to-face interviews, phone interviews, research vertical, site interception, online exploration and systematic observation.

Chosen the appropriate for the research

Based on the case of the AI in reproductive in selection embryos system, we using the quantitative and qualitative research method.

Subject: Computing Research Project

Artificial Intelligence system A combination of different feature extractors and classification methods allows for greater accuracy and reliability. This work proposes a multi-glazing system that combines good structural descriptions with high-performance general purpose classifications. The architecture of the system is mapped in Figure 2. A full description of the preprocessing steps, feature extraction and grading is given in the following sections. Figure 2 Proposed system for embryos and oocytes image classification.

Segmentation and preprocessing

The first step of the pretreatment process is to manually segment the area of interest from the ground. After segmentation, the image is resized into fixed sizes (75 × 75 pixels) and pre-processed by applying a contrast enhancement method (Figure 3), to address issues. inherent heterogeneous lighting. Figure 3 Segmented images (left) and their enhanced versions (right) of an oocyte and an embryo.

Subject: Computing Research Project

Feature extraction

Good texture descriptions are invariant to image rotation and scaling and can be strong on variations in lighting. This study used LBP (Timo Ojala, Matti Pietikäinen, Topi Mäenpää, 2002)a local structural operator with strong discrimination, low computational complexity and low sensitivity to changes in projection. The project has been successfully applied to biological issues.

Classification

This work uses a group of neural networks to perform classification tasks. An artificial neural network is a set of simple processing elements connected together to form a network of nodes that use mathematical models to process information. Different types of neural networks can be obtained by changing the network architecture and choosing algorithms designed to infer the intensity (weight) of the connections in the network to generate the signal flow. desire. A special type of network - the LevenbergTHER Marquest neural network - is used. An independent classifier is not a good option in this classification problem, so this job uses a random subset of classifiers, drawing a subset of all the features available for training classifiers in a group. This system helps partially solve the problem of low number of samples in training. The final score of the orchestra is obtained by calculating the sum of all grades ('total rule').

Statistical analysis

The accuracy of the proposed decision support system depends on the degree to which the tested image group (egg cell / embryo) is divided into two layers in question (results / does not lead to pregnancy). Accuracy is measured by the area under the receiver operating characteristic curve (ROC) (AUC) .Area 1 represents a perfect system; The area 0.5 represents a null system (similar results can be obtained by randomly selecting the output layer). The preliminary guide to assessing the accuracy of a system is the traditional academic scoring system: 0.9 allowed1 = excellent (A); 0.8 balls 0.9 = good (B); 0.7 0.8 = fair (C); 0.6 balls0.7 = poor (D); 0.5 iron0.6 = fail (F). RESULT The purpose of this section is to validate the proposed approach to the existing data set, according to two different test protocols. The first test protocol was a 'woman leaving', using only embryos / oocytes where the label was firm: the test set included all the egg / embryonic cells of a given woman (only consider those with labels that are for sure). Therefore, results were obtained by looking at the performance of 62 experiments (Table 3). The second test protocol is 'single woman' using all women of the data set. Therefore, results were obtained by considering the performance of 104 experiments (Table 3). The delivery rate in this study is 26.5%. It is not affected by the cause of infertility or transplant failure many times. Using the first test protocol, the AUC is calculated according to the score of each embryo / oocyte of the test set, as their labels can be obtained without certainty. Using the second test protocol, classes of images in the test set can be reliably assigned only if all the n-transfer embryos have resulted in no birth or n birth (out of 57). 104 experiments); therefore, AUC is calculated by the score and label of each woman, determined in the following way: for each woman, the maximum score of her embryos / oocytes is selected and her label is only is 'born' if the woman gives birth at least once. The first test aimed to compare different descriptions and classifications to classify both oocytes and embryos. Specifically, in Table 4, Table 5, Table 6, Table 7 two structural descriptions are considered, ie

Subject: Computing Research Project

Pappalardo S., 2012)even if the results were not directly compared because they used a different test protocol and data set. The second test is to evaluate the proposed method using semi-supervised learning method. For this second test, only the last method proposed in this article (LBP + RSNN) is used. The results reported in Table 8 are obtained by using an 'imperfect teacher' to label uncertain training models (Manna et al., 2004). In a nutshell, certain cases of training are used to classify other types of training, that is, if in a group of oocytes / embryos is transferred to the same uterus, then d, then the embryo d has the highest similarity to the 'birth' class assigned to that class and the remaining n - d is assigned to the 'unborn' class. This procedure allows uncertain labels to be removed. Data set Testing protocol First Second Oocytes 0.73 0. Embryos 0.81 0. Table 8 AUC obtained using local binary patterns and random subspace ensemble of neural networks and one iteration of the semi-supervised system. Finally, in Table 9, the proposed method is evaluated using a different semi-supervised learning system, including performing 150 cross-validation 10 times on the training set. 10-fold cross-validation involves randomly dividing the data set into 10 equal-sized subsets D i: nine of these subsets are used for training and one set is a test set. Each time, training samples with uncertain labels are assigned to a certain class and the final labels are taken according to the main voting rule out of 150 times. In any case if a group of n embryos is transferred to the same uterus that gave birth to d, only those embryos with the highest similarity to the 'birth' of the class are assigned to that layer. Data set Testing protocol First Second Oocytes 0.8 0. Embryos 0.83 0. Table 9 AUC obtained using local binary pattern features and 150 iterations of the semi-supervised system DISCUSSION This article focuses on a new method for classifying embryo and oocyte images based on a structural description (local binary sample) and on a random subspace of the Levenberg neural network. The results are clearly superior to existing methods (C Manna1, G Patrizi2,4, A Rahman3, H Sallam3, 2004)and are encouraging, especially considering that they have been obtained using 'small' training with very few positive samples (.80.8 AUC looks at test protocol of neglecting a woman). For oocytes and embryos, other types of descriptors not specifically designed for the common texture image may be used. In fact, the association and the number of nucleoli or the number and size of the embryo are not structural features.

Subject: Computing Research Project

Experience has shown that embryo quality is not clearly related to oocyte appearance. Similar rates of pregnancy and clinical implants have been published after transferring embryos from 'abnormal' or 'normal' oocytes (Balaban, 2011). However, some points based on morphological appearance of oocytes have been proposed to indicate some potential development of subsequent embryos. The features incorporated in the texture of images are often not perceived by the human eye and their analysis by artificial intelligence can be used in a new tool to identify embryos or oocytes. Feasibility. At the same time, human perception of other features (i.e., homogeneity and number of embryos with nuclei, fragmentation, number and treatment of nuclei) is subjective and also limited. obviously because they change over time. A ROC curve for the semi-supervised learning system is depicted in Figure 3. The AUC of this system is about 0.8 using the first test protocol and can be considered as good. It is interesting to note that only supervised classifications are helpful to improve classification performance, perhaps because in this matter only a few samples have a certain 'birth' label. Another interesting result of the experiment is that the best performance is obtained when using egg cells: this is due to the fact that the egg cells are more similar in structure, so they may be more suitable for Structural information analysis. Table 10 Receiver operating characteristic (ROC) curves: (A) oocytes with first testing protocol (1TP), (B) oocytes with second testing protocol (2TP), (C) embryos with 1TP, (D) embryos with 2TP. Areas under the curve are reported in Table 9.