Aristotle University Of Thessaloniki (AUTH)
Δημιουργία διαλειτουργικού πλαισίου για την ενοποιημένη πρόσβαση σε ετερογενή πολυμεσικά ιατρικά δεδομένα
Abstract
dc:descriptionThe research scope of this particular thesis, focused in the study, representation, management and integration of biosignals, in the cardiology domain. ECGs constitute cardiology domain biosignals, which are traditionally store, either in electronic data formats, which are mainly proprietary, or converted to widely accepted ECG coding standards like EDF (European Data Format), SCP-ECG (Standard communications protocol for computer assisted electrocardiography) etc. Apart from the signal itself, most of these standards encompass the equally important context and content meta-data of the signal. Biosignal context refers to information about patient demographics, diagnosis, recording equipment, researcher/investigator, etc while annotation and interpretation information constitute part of the bio-signal content. This thesis was motivated by the requirement of an integrated and interoperable environment, for the unified study of mainly heterogeneous biosignals, originating from different population groups and recorded under diverse conditions. Interoperability, defined as a prerequisite in this work, introduced the notion of standards, which are numerous in the healthcare domain. Most of the standards encode ECGs in flat file formats, which lack the information necessary to support meaningful analysis, interoperability and integration of multiple resources. To address the above mentioned issues, knowledge representation technologies were studied, the selection of which typically constitutes a critical decision step during data integration process. Ontologies were chosen as the representation structure, since they enable the reuse of domain knowledge without reinventing the wheel. Furthermore, ontologies constitute the backbone of the semantic web and can be easily comprehensible by humans and machines. Next, data integration approaches were studied resulting in the definition of a biosignal integration methodology, which employed the conceptual data integration model and the Local As View approach. In the scope of the proposed data integration methodology, the global ontology was modeled according to ecgML mark up language hierarchy, which encompasses the structure and content of the most widely accepted standards in the cardiology domain. Additionally ontologies were introduced, not only as structures for the virtual unification of the underlying sources but also as platforms for the transparent content and context based searching from multiple data sources. The decoupling of the content and context of the signal was also proposed, in order to achieve its semantic enrichment and enable structured and comprehensible searching. Applying the above mentioned methodology to heterogeneous sources of ECG signals, induced the development of ROISES (Research Oriented Integration System of ECG Sources), which aimed at the semantic integration of the sources and enabled unified content and context base searching. The global ontology, which constituted the system’s search structure, enabled the clinician to follow alternative diagnostic and therapeutic pathways that could possible lead to hidden knowledge and unexpected results. The clinician could easily express queries constructed upon a structured terminology and redefine the relations among concepts. Such concepts could emerge after the meta analysis of the therapeutic procedure, based on diagnosis and the possibility of reevaluation of diagnosis and finally its mapping to standardized clinical guidelines. The next step included studying of ontology learning approaches, in order to develop a new ontology enrichment method, designed specifically for ROISES. The resources that were encapsulated to the system consisted of ECG collections encoded in EDF and SCP-ECG standards and a relational database for the management of patients with chronic diseases. Leveraging ROISES with new resources and resources with more data, induced the need for the semantic enrichment of the global ontology. Dynamic Terminology Enhancement Method (DTEM) was developed to address this issue, based on UMLS Metathesaurus services and machine learning techniques. Research and application of different machine learning techniques like Support Vector Machines and decision tree learning algorithms lead to the selection of the ones that rendered the most efficient results pertaining accuracy, specificity and sensitivity. DTEM manipulated the medical terms that were not mapped to any concepts in the global ontology, to primarily determine their suitability as potential classes in the global ontology hierarchy and secondarily, given the suitability of a medical term, elucidate its location in the hierarchy. The innovation of this study was related to both the approaches proposed and their application domain. More specifically, the proposed methodology lead to the implementation of a unified interoperable framework for studying and managing biosignals, by employing medical technology and medical terminology standards, both defined as ontologies. The global ontology, designed in a prototype hierarchical structure, provides alternative diagnostic and therapeutic pathways to the medical researcher that may lead to unexpected results. Concerning the application domain, semantic data integration has not been applied to biomedical signals and particularly ECG’s. This thesis substantially contributed to the dynamic semantic enrichment of ROISES data integration system, presented in chapter 5. The proposed approach enabled the dynamic enrichment of the global ontology with concepts originating from the resources. In this scope this approach, based on machine learning and UMLS Metathesaurus, added new concepts in the taxonomy of the global ontology in a semi-automatic manner. Additionally it contributed to the enrichment of the system’s search criteria in order to study the various parameters of ECG biosignals, revealing implicit knowledge and promoting research in the cardiology domain. Adding new data into existing resources and new resources encoded in various standards apart from SCP-ECG and EDF, constitutes a future step towards the system’s extension. Applying slight changes to the structure and terminology of the global ontology, could also allow for the encapsulation of other biomedical signals like EEG. Finally, adding real data, medical guidelines and therapeutic pathways will foster the system’s capability for medical decisions support.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2011
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Kokkinaki, Alexandra
- Κοκκινάκη, Αλεξάνδρα
Subjects
dc:subject × 12- Σημασιολογική ενοποίηση ηλεκτροκαρδιογραφημάτων
- Σημασιολογική αναζήτηση ηλεκτροκαρδιαγραφημάτων
- Σύστημα ολοκλήρωσης ΗΚΓ δεδομένων
- Οντολογικός εμπλουτισμός
- Semantic integration of ECG signals
- Semantic searching of ECG signals
- ECG integration system
- Ontology enrichment
- Φυσικές Επιστήμες
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Natural Sciences
- Computer and Information Sciences
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/33155