Skip to content. | Skip to navigation

Informatik 5
Information Systems
Prof. Dr. M. Jarke
Sections
Personal tools
You are here: Home Staff Dr. Oya Deniz Beyan

Dr. Oya Deniz Beyan - Theses


Bachelor
Data Quality for Fitness of Use - Running
Linked data is gaining new attention in the last years because of its natural connection to knowledge-based applications. The quality of decisions depends heavily on the quality of the underlying data, for reasoning such quality reports are mandatory for each decision. The W3Cs Best Practices Working Groups "Data on the Web Best Practices: Data Quality Vocabulary" defines a vocabulary to archive linking results of data quality assessments to linked data. Also, a basic set of quality dimensions and metrics based on the work of Zaveri et al. (https://dx.doi.org/10.3233/SW-150175) are presented. This thesis aims to fill the gaps between the DQV, the definitions by Zaveri et al. and the realization of linked data quality assessments, to fulfil all requirements to link data quality assessments.
Master
Unsupervised Anomaly Detection in Medical Time Series Data - Finished
Completed by Brockhoff, Tobias in 2019; Supervised by Prof. Dr. Stefan Decker; Advisor(s): Dr. Oya Deniz Beyan
Master
Deep Learning-based Knee Osteoarthritis Diagnosis from Radiographs and Magnetic Resonance Images - Finished
Completed by Jiao, Jiao in 2019; Supervised by Prof. Dr. Stefan Decker; Advisor(s): Dr. Oya Deniz Beyan, Dr. Michael Cochez, Md. Rezaul Karim, M.Sc.
In this thesis, the student investigates the use of deep learning techniques (especially computer vision) to perform diagnosis of osteoarthritis. The input to the system are both radiographs (X-RAY) and magnetic resonance images (MRI).
Bachelor
Classification of Cancer with methylation aware motifs - Finished
Completed by Gehrmann, Julia in 2019; Supervised by Prof. Dr. Stefan Decker, Ivan Gesteira Costa Filho; Advisor(s): Dr. Oya Deniz Beyan, Md. Rezaul Karim, M.Sc.
This dissertation addresses the problem of classification of cancer patients from DNA methylation. Mrs. Gehrman explores here the use of scores of transcription factors binding sites around DNA methylation as surrogate markers for DNA methylation. An innovative aspect is the fact binding site motifs take into consideration of the DNA methylation status of a given locus. Next, this work compared the performance of machine learning classifiers either using classical DNA methylation levels vs. TF binding affinity scores. For this, several classical machine learning methods (SVM, inductive trees, random forests) were used. Performance accuracy was similar with both data representations, however the computational time of training classifiers with TF binding site affinity scores were at least 10 times fasters, due to the lower dimensionality of the new space. This works supports promising features of DNA methylation aware TF binding scores.
Master
Master
Developing a machine-actionable process model for ethical approval workflows - Finished
Completed by Schwaiger, Verena in 2019; Supervised by Prof. Dr. Stefan Decker; Advisor(s): Dr. Oya Deniz Beyan
Bachelor
Master
Patterns zur Integration regel- und prozessbasierter Modellkomponenten bei der Operationalisierung klinischer Leitlinien - Finished
Completed by Lammers, Florian in 2019; Supervised by Prof. Dr. Stefan Decker, Dr. rer. nat. Cord Spreckelsen; Advisor(s): Dr. Oya Deniz Beyan
Master
An Explainable and Fully Automated Localization and Semantic Segmentation Technique for Biomedical Imaging - Finished
Completed by Shafin, Nazmi in 2019; Supervised by Prof. Dr. Stefan Decker; Advisor(s): Dr. Oya Deniz Beyan, Md. Rezaul Karim, M.Sc.
Timely and correct intervention with surgical procedures by physicians is the most critical factor in reducing the morbidity rate of patients with chronic brain ailments, in particular patients with brain Tumor. Deep learning techniques in computer vision is making rapid progress in diagnosing and thereby aiding the physician in taking timely appropriate steps, and thus it has the potential to save many lives. However, often the deep learning models are like a black box that merely generates the outcome without giving any reason or explanation. The field of Artificial Intelligence needs models that have a capability to include explainability of the decisions that it is making. The US Food and Drug Administration requires any clinical decision support software to explain the rationale or support for its decisions to enable the users to independently review the basis of their recommendations. To meet this requirement and gain trust from medical practitioners, medical deep-learning systems should provide explanations for their outputs. This thesis work is an endeavour in that direction of explainable artificial intelligence whereby the results are generated and supplemented with an explanation of the decision that can be interpreted with visualizations generated by localization and semantic segmentation techniques which will ultimately be of significant help in aiding physicians for treating critical ailments such as brain tumor.
Master
FAIR Identifier Registry for Distributed Systems - Finished
Completed by Snizhko, Oleksandr in 2019; Supervised by Prof. Dr. Stefan Decker; Advisor(s): Dr. Oya Deniz Beyan, Heiner Oberkampf, Jan Winkelmann
The active increase of digital data, as well as the generation of new varieties of digital content, has established several possibilities and challenges in the management of the big amount of data. Traditionally, organisations have relied on Uniform Resource Locator (URL) hyperlinks to provide involved parties with access to their digitised content via the Internet. Nevertheless, over time, more of these hyperlinks become invalid. The concept of persistent identification has been developed to solve this issue. Instead of addressing data directly through its actual locator, Persistent Identifiers (PIDs) permit data retrieval by globally unique and permanent identifiers. PIDs include metadata and resolving URLs that point to the original location of data collections. As the target URLs are changing, PIDs require constant maintenance. Existing PID systems has enabled long-term stable and unambiguous references. They are already in rapid use, supporting distributed approaches at varying levels. In most of the cases, PIDs are indexed and explored by a central system. Besides infrastructural questions, the discoverability of PIDs mainly depends on their metadata classifications, which express what a given PID represents. The aim of this thesis is to explore the creation of a meta-standard for the specification of domain-specific PID registries which are exposed by an application programming interface (API), while standard and API both respect Findable, Accessible, Interoperable and Reusable (FAIR) principles.
Bachelor
Towards Automating Graph Data Cleansing Using Shapes - Finished
Completed by Eibl, Erhard in 2019; Supervised by Prof. Dr. Stefan Decker; Advisor(s): Dr. Oya Deniz Beyan, Heiner Oberkampf, Jan Winkelmann
Data curation is a labour intensive process transforming often completely unstructured data into structured data. This work proposes an approach to improve the quality of semi structured data efficiently and resolve conflicts with minimal effort. This is achieved by inferring a data model or improving a given one, by finding inconsistencies within the data, and by suggesting possible data edits using different machine learning and data mining techniques. Effort is further reduced by detecting conflicts that were caused by other conflicts using root cause analysis.
Master
Semantic Anomaly Detection in Medical Time Series - Finished
Completed by Sven Festag in 2019; Supervised by Prof. Dr. Stefan Decker, Dr. rer. nat. Cord Spreckelsen; Advisor(s): Dr. Oya Deniz Beyan
Master
Using Deep Neural Networks and Copy Number Variations for Cancer Detection - Finished
Completed by Rahman, Md Ashiqur in 2018; Supervised by Prof. Dr. Stefan Decker; Advisor(s): Dr. Oya Deniz Beyan, Md. Rezaul Karim, M.Sc.
Recently copy number variations got lot of attention due to their association with complex diseases like cancer. Finding their association with cancer is an active research eld. We propose a method to identify copy number variations which might be associated with a cancer by comparing copy numbers across 14 major cancer types. Selected CNV regions were used as features for cancer detection using deep belief networks, a deep learning technique. To our knowledge this is the rst study involving that many cancers and highest number of control samples for validation.
Person Publications

Peter McQuilton, Dominique Batista, Oya Beyan, Ramon Granell, Simon Coles, Massimiliano Izzo, Allyson L. Lister, Robert Pergl, Philippe Rocca-Serra, Ben Schaap, Hugh Shanahan, Milo Thurston, Susanna-Assunta Sansone

Helping the consumers and producers of standards, repositories and policies to enable FAIR data

Data Intelligence

Oya Beyan, Ananya Choudry, Johan van Soest, Oliver Kohlbacher, Lukas Zimmermann, Holger Stenzhorn, Md. Rezaul Karim, Michel Dumontier, Stefan Decker, Luiz Olavo Bonino da Silva Santos, Andre Dekker

Distributed Analytics on Sensitive Medical Data: The Personal Health Train

Data Intelligence

Md. Rezaul Karim, Galih Wicaksono, Ivan G. Costo, Stefan Decker, Oya Beyan

Prognostically Relevant Subtypes and Survival Prediction for Breast Cancer Based on Multimodal Genomics Data

IEEE Access

More publications…