Методи штучного інтелекту для аналізу патогістологічних зображень для визначення раку молочної залози
| dc.contributor.advisor | Іванько, Катерина Олегівна | |
| dc.contributor.author | Вознюк, Тарас Русланович | |
| dc.date.accessioned | 2023-08-30T08:21:54Z | |
| dc.date.available | 2023-08-30T08:21:54Z | |
| dc.date.issued | 2021 | |
| dc.description.abstract | Об’єктом розгляду роботи є пухлини раку молочної залози. Предмет роботи – аналіз гістопатологічного зображення ядер пухлини, отриманих за допомогою мікроскопа. Метою роботи є розробка алгоритму класифікації типу пухлин з використанням методів машинного навчання. Перший розділ включає загальний огляд захворювання раку молочної залози, класифікацію, опис типів пухлин та їх відмінності, стадії розвитку та фактори ризику. Другий розділ містить інформацію про будову та принцип дії оптичного мікроскопу та існуючі методи мікроскопії. У третьому розділі аналізуються відомі методи цифрової обробки гістопатологічних зображень, а також описується їх алгоритм дії та загальне призначення. Четвертий розділ складається з аналізу методів класифікації машинного навчання, їх порівнянна, принцип дії та доцільність використання П’ятий розділ полягає у розробці алгоритму класифікації типу пухлин з використанням методів класифікації машинного навчання та конструювання додатку у вигляді графічного інтерфейсу для виконання задачі класифікації. | uk |
| dc.description.abstractother | The object of consideration of the work is a tumours of breast cancer. The subject of the work is the analysis of histopathological image of tumor nuclei obtained by microscope. The purpose of the work is to develop an algorithm for classifying the type of tumors using machine learning methods. The first section includes a general overview of disease of breast cancer, classification, description of tumor types and their differences, stages of development and risk factors. The second section contains information about the structure and principle of operation of an optical microscope and existing methods of microscopy. The third section is about the known methods of digital processing for histopathological images and it also describes their algorithm and general purpose. The fourth section provides information on the application methods of classification of machine learning, their comparsion, principle of operation and expediency of use. The fifth section describes the development of a machine learning algorithm, which consists of image preprocessing, classification the type of tumors and designing an application in the form of a graphical interface to perform the classification task. The purpose of section 1 is to provide an overview of the nature of breast cancer, risk factors that contribute to its occurrence, stages of development, types of tumors, their differences and classification. Cancer is a disease that occurs when the normal cycle of cells in the body is disrupted. Cells begin to divide faster than the body requires, and the old ones do not die, resulting in a tumor. Normal cells can become cancerous after going through such stages as hyperplasia and cell dysplasia. With hyperplasia, there is an increase in the number of cells in an organ or tissue that appears normal under a microscope. Dysplasia disrupts the cell structure (by the increase in the size of the nucleus), which is a prerequisite for cancer. Cancer itself is a malignant tumor and its ability to spread to other parts of the body, thus creating metastases when the benign tumor is fixed in size and can be removed surgically. Breast cancer is dangerous mainly for women and occurs in such parts of the gland as ducts or lobules, based on which its main types are defined. The main risk factors include radiation exposure in adolescence, heredity, past cancer treatment and lack of a healthy lifestyle. Among the stages of development there can be identified 3-4 the most critical ones, when cancer cells begin to metastasize and treatment becomes very difficult. The purpose of section 2 is to consider the main stages of preparation of the sample of the affected tissue for analysis under a light microscope, the structure of which is described in detail, and the most common types of microscopy, which have their own characteristics. One of the most important reasons for high-quality and accurate diagnosis of pathology is the preliminary controlled treatment of the test sample with special solvents and fixing it on a glass slide, which will help specialists to recognize the signs of the disease better. Pathology, histopathology or histology is aimed at studying the manifestations of the disease by microscopic examination of tissue morphology. In case of pathology, the test specimen is subject to surgery, biopsy or dissection after fixation, cleaning and cutting of the specimen. Tissue sections after wax fixation and surfacing are typically cut into thin sections of two to five microns before staining and transferred to a glass slide for examination with a light microscope. Preparation of histopathological slide begins with the fixation of a tissue sample, the purpose of which is to prevent tissue rot. For best results, biological tissue samples should be fixed immediately after collection, usually in 10% neutral formalin for 24-48 hours. After fixation, the samples are cut with a scalpel to allow them to fit into a suitable labeled fabric cassette, which is stored in formalin prior to processing. The first step of treatment is dehydration, which involves immersing the sample in increasing concentrations of alcohol to remove water and formalin. Purification is the next step in which an organic solvent is used to remove the alcohol and ensure infiltration with paraffin wax. The microtome is used to cut extremely thin sections of tissue from a block in the form of a tape, after histochemical staining, to provide contrast with areas of tissue, making tissue structures more visible and easier to assess. When considering some methods of microscopy, we can distinguish several promising ones, and namely phase-contrast microscopy, dark field microscopy and confocal microscopy. Section 3 discusses the most commonly used methods of digital processing of histopathological images, such as gamma - correction, deconvolution of the image, color schemes, different methods of noise filtering, histogram brightness and its alignment, area of interest, different types of edge detectors (among which the most appropriate ones are detectors Sobel and Canney), as well as describes the known types of segmentation of cells and nuclei on a histopathological sample. Section 4 discusses the topic of machine learning, and namely the general purpose, evaluation of the model effectiveness, the main stages of machine learning from data preparation to testing of the trained model, and the method of reducing the dimensionality of features (PCA). Machine learning is a branch of artificial intelligence that uses the techniques of automatic learning and improvement of systems based on experience. Machine learning focuses on developing computer programs that can access data and use it for self-study. Machine learning algorithms are often classified as learning with or without a teacher. Machine learning algorithms with a teacher can apply things that have been learned in the past to new data, using labeled examples to predict future events. Starting from the analysis of a known set of training data, the learning algorithm produces a derived function for predicting the initial values. The learning algorithm can also compare its results with the correct, predictable result and find errors to modify the model accordingly. In contrast, non-teacher machine learning algorithms are applied when the information used for learning is not classified or labeled. Unattended learning explores how systems can infer a function that describes a hidden structure out of unlabeled data. Machine learning allows you to analyze a huge amount of data. This provides faster and more accurate results for identifying profitable opportunities or dangerous risks, but it may require additional time and resources to prepare properly. The combination of machine learning with artificial intelligence and cognitive technology can make it even more effective in processing large amounts of information. There are also described in detail such teacher training methods as logistic regression method, linear discriminant analysis, nearest neighbors method k - NN, SVM reference vectors method, random tree of decision tree method and ensemble methods. In Section 5, good results were achieved in the implementation of the task of classifying the type of breast tumors (benign or malignant) for the studied images of histopathological samples using machine learning methods. Several image processing methods were applied to digital images of the studied histopathological samples, which made it possible to achieve high accuracy in calculating the characteristics of cancer cell nuclei for further use in the classification problem. The features of tumor cell nuclei were calculated and a vector of features was created. Using the Breast Cancer Wisconsin (Diagnostic) Data Set for Machine Learning, the best model for classifying breast tumors into 2 classes (benign or malignant) was obtained – and namely, a model of the reference vector method with an overall accuracy of 96.2% and optimal class accuracy compared to others. The most important efficiency metrics such as accuracy, area under the curve (AUC), F1-measure and mismatch matrix were calculated for the model. There was provided an example of obtaining a diagnosis for the studied pathohistological sample of breast tumor - label "M", which indicates the presence of a malignant tumor. There was developed an own application with a graphical interface for convenient and interactive execution of the algorithm of actions from sample loading and image pre- processing to direct classification of the tumor using a trained model with the highest accuracy. | uk |
| dc.format.extent | 97 с. | uk |
| dc.identifier.citation | Вознюк, Т. Р. Методи штучного інтелекту для аналізу патогістологічних зображень для визначення раку молочної залози : дипломна робота … бакалавра : 153 Мікро- та наносистемна техніка / Вознюк Тарас Русланович. – Київ, 2021. – 97 с. | uk |
| dc.identifier.uri | https://ela.kpi.ua/handle/123456789/59651 | |
| dc.language.iso | uk | uk |
| dc.publisher | КПІ ім. Ігоря Сікорського | uk |
| dc.publisher.place | Київ | uk |
| dc.subject | рак молочної залози | uk |
| dc.subject | цифрова обробка зображень | uk |
| dc.subject | класифікація | uk |
| dc.subject | машинне навчання | uk |
| dc.title | Методи штучного інтелекту для аналізу патогістологічних зображень для визначення раку молочної залози | uk |
| dc.type | Bachelor Thesis | uk |
Файли
Контейнер файлів
1 - 1 з 1
Вантажиться...
- Назва:
- Vozniuk_bakalavr.pdf
- Розмір:
- 3.74 MB
- Формат:
- Adobe Portable Document Format
- Опис:
Ліцензійна угода
1 - 1 з 1
Ескіз недоступний
- Назва:
- license.txt
- Розмір:
- 9.1 KB
- Формат:
- Item-specific license agreed upon to submission
- Опис: