Ing. Jakub Žitný

Ing. Jakub Žitný

Ing. Jakub Novák

Department of Applied Mathematics

This thesis contains research of the field of medical imaging, classical methods of image segmentation, computed tomography and convolutional neural networks. The practical part involves implementation of an architecture of 3D UNet for segmentation of the spine and specific vertebrae from CT scans. Furthermore, this architecture is compared to its 2D counterpart.

Thesis on DSpace

Ing. Jakub Žitný

Ing. Viktor Černý

Department of Software Engineering

This work deals with analysis of backend of LearnShell system, proposal of improvements and their implementation. The main proposed improvement is a score bonuses and penalties system, which was developed by implementation of a domain-specific language. The logic of applying of bonuses and penalties can be programmed for by using this language. In this work, I also analyse pricing of some major cloud providers, where the LearnShell will migrate in the future.

Thesis on DSpace

Ing. Jakub Žitný

prof. RNDr. Pavel Surynek, Ph.D.

Department of Applied Mathematics

This thesis focuses on the research of techniques used for the detection and classification of defects on the CT images using neural networks. Based on these methods, the own implementation, which works for intracranial hemorrhage detection on CT images of the brain, is described. This solution is based on convolutional neural networks, and the emphasis is primarily on the data preprocessing and the selection of the appropriate architecture and its hyperparameters. The final model achieved accuracy, AUC, sensitivity, and specificity of 0.963, 0.967, 0.840, and 0.970, respectively. This result is a good basis for further research and can be additionally improved by other experiments.

Thesis on DSpace

Ing. Jakub Žitný

Ing. Jiří Kašpar

Department of Software Engineering

The main goal of this thesis is to present a functional cluster infrastructure for the LearnShell 2.0 project used by the Czech Technical University in Prague. To achieve this, we have described the current infrastructure and scrutinized today's pre-eminent technologies for container orchestration. As a result, we present a basic cluster on which the LearnShell project is to be hosted, and have also defined CI/CD routines for the services contained therein by utilizing Gitlab CI. We have documented our code thoroughly and display our code in a separate repository.

Thesis on DSpace

Ing. Jakub Žitný

Ing. Magda Friedjungová

Department of Applied Mathematics

Diabetic retinopathy is a serious eye disease that can lead to total blindness. Early diagnostic significantly increases the chances of successful treatment of it. In this bachelor's thesis, we built a convolutional neural network to classify photos of eye fundus into 5 stages of diabetic retinopathy. Also, we introduced other studies related to automatic diabetic retinopathy diagnostic and the theory behind it.

Thesis on DSpace

Ing. Jakub Žitný

Ing. Magda Friedjungová, Ph.D.

Department of Applied Mathematics

The aim of this work is to try to apply the recently proposed generative adversarial network model for the classification and segmentation of medical images and to compare the results with modern GAN models currently used for medical data augmentation.

Thesis on DSpace

Ing. Jakub Žitný

Ing. Magda Friedjungová

Department of Applied Mathematics

Improving and developing new algorithms in the field of machine learninghas impact on everyday life. As the performance and size of datasets increase,tasks that were previously considered unattainable metas become comprehen-sible and manageable. Object segmentation is one of the most common tasksof computer vision today. It is widely used in car technology, industry, tradeand healthcare.This work aims to design, train and compare the performance of the modelwith other State-of-the-art models within the segmentation of CT scans of pa-tients. It provides a theoretical basis and acquaints the reader. In this work,great emphasis is placed on the overall process of development of machinelearning models. It turns out that the prototype of the proposed neural net-work achieves better results than neural networks with a larger number ofparameters.

Thesis on DSpace

Ing. Jakub Žitný

doc. Ing. Kamil Dedecius, Ph.D.

Department of Applied Mathematics

This thesis aims to summarize different improvement techniques and state of the art approaches usable for tasks in medical imaging domain. With this knowledge, convolutional neural network models are implemented and and trained on a musculoskeletal radiographs dataset. The impact on model performance by applying multiple preprocessing methods is watched and dis- cussed.

Thesis on DSpace

Ing. Jakub Žitný

Ing. Viktor Černý

Department of Software Engineering

In this bachelor's thesis my main focus is on improving the application: LearnShell~2.0. My main goal is to add an analytics module to the application. This module calculates statistical data which can then be visualised in the web application. Another part of my thesis is a financial analysis of cloud service providers ideal for migrating LearnShell to, in the future. My contribution is making LearnShell better for students resp. teachers who use this application -- making it more convenient to teach resp. study.

Thesis on DSpace

Ing. Jakub Žitný

Ing. Lukáš Bařinka

Department of Software Engineering

This thesis is about designing, implementing, and testing a new solution for exam management in the LearnShell application. The current solution lacks the user interface for creating and managing exams. The solution started with an analysis of the backend and similar applications. The next step was creating a design, improving the tooling used to program the frontend, and implementing the solution. Finally, the application was tested both by usability testing and automatic tests. The result is a functioning system for exam management, which will be used in the Faculty of Information Technology CTU in Prague.

Thesis on DSpace

Ing. Jakub Žitný

Ing. Jiří Kašpar

Department of Software Engineering

The following thesis is about analysis of current LearnShell system that is used in current education process on our faculty. In the text below I'll have a look on the current situation, observe all existing bottlenecks, propose solutions for some of them and define a roadmap of what are the next steps that can be done here. Also, I'll explain selected approaches and technology stack and compare it to the current market trends.

Thesis on DSpace

Ing. Jakub Žitný

Ing. Milan Dojčinovski, Ph.D.

Department of Computer Systems

This bachelor’s thesis describes the development of a module for generating GraphQL API for the Django web framework. The main part of this thesis is concerned with furthering the development of Simple API module, which aims to facilitate fast creation of API endpoints from existing Django definitions. The thesis contains an analysis of existing modules for API creation in Django, not just GraphQL, but also of those generating REST-style API. From the analysis of the current state of Simple API development, new additions to Simple API are designed, then implemented. Other than functional additions, the development also concerns the design and implementation of security features necessary for prevention of Denial of Service (DoS) and batching attacks through complex queries. Furthermore, a web application for testing of the API is also developed in the thesis. The result of this thesis is the expansion of the module Simple API and web application Simple API admin for exploration and testing of the generated endpoint.

Thesis on DSpace

Ing. Jakub Žitný

Ing. Jiří Kašpar

Department of Software Engineering

This thesis focuses on developing a tool for communication between the Grades assessment system and the LearnShell application. LearnShell, being an independent educational platform, manages all its data internally. As a result, none of the LearnShell classification data is accessible via Grades. Moreover, this problem is not unique to LearnShell. Design and implementation of a module allowing easy integration of Grades into any Node.js-based application is presented. The proposed design of the module is based on the research and analysis of Grades API and relevant web technologies. The implemented module is further subjected to unit testing; it is also integrated into a test application. The developed module is publicly available and freely obtainable via NPM. A manual describing the process of installation, setting-up, and usage is included.

Thesis on DSpace

Ing. Jakub Žitný

Ing. Tomáš Kalvoda, Ph.D.

Department of Applied Mathematics

The current approach to the diagnosis and quantification of the joint damage caused by rheumathoid arthritis is a manual radiographic image inspection by a radiologist, which is generally expensive, time-consuming and subjective. Automated assessment systems are a way to overcome this problems and introduce more objectivity into radiology reports, coupled with a faster damage quantifying. The possibility of such a system development will be shown on the example of a participation in RA2 DREAM Challenge. As a part of the competition, the system utilizing several state-of-art convolutional neural net architectures will be developed and scored on the University of Alabama at Birmingham Cheaha supercomputing system.

Thesis on DSpace

Ing. Jakub Žitný

doc. Ing. Kamil Dedecius, Ph.D.

Department of Applied Mathematics

The COVID-19 pandemic is a very pressing issue that continues to affect the lives of people around the globe. To combat and overcome the disease, it is necessary for infected patients to be quickly identified and isolated to prevent the virus from spreading. The traditional detection techniques based on molecular diagnosis, such as RT-PCR, are expensive, time-consuming, and their reliability has been shown to fluctuate. In this thesis, we research the detection of COVID-19 in chest X-ray images using convolutional neural networks. We use our findings to implement a prototype that performs binary detection of the disease, evaluate its performance on a collection of open data repositories available online, and compare its results to existing models. Our proposed light-weight architecture called the BaseNet achieves an accuracy of 95.50 % on the chosen test set, with a COVID-19 sensitivity of 93.00 %. We further assemble an ensemble of the BaseNet along with several other fine-tuned architectures, whose combined classification accuracy is 99.25 % with a measured sensitivity of 98.50 %.

Thesis on DSpace

Ing. Jakub Žitný

Ing. Tomáš Kalvoda, Ph.D.

Department of Applied Mathematics

X-ray is one of the cheapest and most popular diagnostic methods. Automating such a task can help doctors assess proper diagnosis. This thesis is about the automatic detection of abnormalities from radiographs.

Thesis on DSpace

Ing. Jakub Žitný

doc. Ing. Kamil Dedecius, Ph.D.

Department of Applied Mathematics

This thesis deals with the insufficient volume of data in the domain of medical imaging related to the covid-19 virus, which hinders proper classification of this severe illness. The main focus is generation and comparison of X-ray images of human lungs. It explores ways to tackle the issue of small number of positive cases in the pathology of covid-19, by utilising, evaluating and comparing different generative models to increase the number of high quality and credible images usable in machine learning algorithms. It compares models of the generative adversarial network family and assesses the impact of using images generated in this manner on the training of classification neural networks. The thesis is aiming to bolster existing classification algorithms to perform better in determining covid-19 infection based on lung X-ray images.

Thesis on DSpace

Ing. Jakub Žitný

Ing. Lukáš Brchl

Department of Applied Mathematics

This thesis summarizes state-of-the-art techniques for prediction, classification, and segmentation tasks, specifically neural networks. The main focus on convolutional neural networks for image classification. The practical part has the implementation of convolutional neural network models on binary classification task for x-ray images of the body parts and subsequent comparison of results with existing models.

Thesis on DSpace

Ing. Jakub Žitný

doc. Ing. Štěpán Starosta, Ph.D.

Department of Applied Mathematics

This work apprises the reader with the basics of machine learning. Furthermore researches the current state-of-the-art approaches used in machine learning in medical imaging. As a part of the theis, the prototype for classification of MURA dataset is trained.

Thesis on DSpace

Ing. Jakub Žitný

Department of Software Engineering

This bachelor's thesis focuses on improving the current module for detecting plagiarism in LearnShell. The goal is to build a front-end React application with an emphasis on user experience to allow teachers to inspect the plagiarism suspicions and resolve them. Another part of this work aims to improve the plagiarism detection algorithm itself. This work describes the whole process of developing such application from analysis and practical examples and descriptions of the implementation of all the laid down goals to some additional theoretical options for future improvements.

Thesis on DSpace

Ing. Jakub Žitný

Department of Software Engineering

LearnShell is a web application for the automatic correction of shell assignments with the missing searching functionality. This thesis deals with the analysis, design, implementation, and testing of a new smart searching module for the system. The goal has been successfully achieved. The module — which has been built upon an open-source search engine Elasticsearch — supports full-text, real-time, incremental search, bool operators, and several types of queries.

Thesis on DSpace

Ing. Jakub Žitný

Ing. Jiří Kašpar

Department of Software Engineering

This thesis deals with the design and development of an analytical module for the LearnShell system for students and teachers. First, the possibilities are analyzed and then a solution is proposed using functional and non-functional requirements and wire frames. The second main part then describes the specific outputs of the module, its implementation and analysis. The output is a module that informs users using educational dashboards about current relevant information and about the predicted performance of students using statistical prediction.

Thesis on DSpace

Ing. Jakub Žitný

Mgr. Jan Spěvák, Ph.D.

Department of Applied Mathematics

This thesis concerns research in the field of computer vision and convolutional neural networks. The practical part contains an implementation of face detection, recognition and customized object detection for legitimization checks. Automation of these checks replaces the standing manual checking and allows deploying them in larger scenarios.

Thesis on DSpace

Ing. Jakub Žitný

Ing. Magda Friedjungová, Ph.D.

Department of Applied Mathematics

Skin cancer is now one of the most common types of cancer. A dermatoscope is a tool, which takes the dermoscopic images of the skin. These images can capture details that can escape the human eye. These images are then evaluated, in many cases, by two doctors, one who determines the diagnosis and the other who checks it (determines his diagnosis). Nowadays, this role of the examining physician can be replaced by precise classification models, which need a large amount of data, which is unfortunately lacking. This work examines the use of deep generative models in the generation of dermoscopic images, which would help solve the mentioned problem of lack of data. With the generative model StyleGAN2ADA, the classification model was improved in the recognition of two types of skin diseases NV and MEL. Class sensitivities improved by 4.29 \% and 0.72 \% compared to using classical data augmentation.

Thesis on DSpace

Ing. Jakub Žitný

Ing. Miroslav Čepek, Ph.D.

Department of Applied Mathematics

The increasing complexity of the state-of-the-art machine learning models has caused them to become opaque black boxes. This hinders their deployment in critical domains such as healthcare. This thesis studies the field of explainable artifical intelligence in the context of medical imaging. It selects and detailedly describes various state-of-the-art explanation methods. The explanation methods are used to explain and evaluated on ResNet50 and Visiontransformer models trained for the task of covid-19 pneumonia detection from chest X-ray scans.

Thesis on DSpace

Ing. Jakub Žitný

doc. Ing. Pavel Kordík, Ph.D.

Department of Applied Mathematics

Github is a place where developers work on projects and share their work with others. Over time, the repository has become the place with the largest code-base in the world. Anyone who decides to solve a problem, create a package or a library can do so and can contribute. This has led to a plethora of new projects that are pushing the boundaries of technology. Gradually, as the technology matures, it becomes more widely known to developers. The solution starts to be integrated into other libraries and other projects integrate parts of the solution. Developers developed it in the good faith that the community will continue to use and support it. Over time, as alternatives were developed, the solution became obsolete and developers started to abandon it. The outflow of people resulted in a decrease in code quality and the gradual disappearance of the project. This work aims to analyse information about repositories in order to determine the quality of the repository and its lifespan.

Thesis on DSpace