http://trepo.tuni.fi:80/handle/10024/105882 fi=Pro gradut, Diplomityöt, Syventävät työt (lääketiede), Lisensiaatintyöt|en=Master's theses, Master's theses (medicine), Licentiate theses| 2026-07-18T09:15:06Z http://trepo.tuni.fi:80/handle/10024/238938 A Multi-Resolution Deep Learning Framework for Seizure Detection in Multi-Channel Electroencephalography Yener, Deniz 2026-01-01T00:00:00Z http://trepo.tuni.fi:80/handle/10024/238911 PointPerfect Atmospheric Corrections Modeling Gergis, Andrew The thesis examined atmospheric correction errors, in particular the ionospheric delay, for the PointPerfect PPP-RTK GNSS service as well as the reliability of the service quality indicator QI of u-blox receivers in different scenarios. For the modeling of the ionospheric residuals, after applying PointPerfect corrections, the dual-frequency observation models ionosphere-free and geometry-free were used. The investigations were performed with multi-GNSS data of the RINEX type as well as with precise products of various IGS stations in Germany and Italy. After downloading the RINEX and precise product files with a MATLAB-based processing stream the corrections, aligned to the SPARTN system, as well as an external GNSS solver were run. From the large datasets of observations, the ionospheric residuals and various quality measures were extracted. The ionospheric residuals were analysed by error detection and by a fit of the errors to so-called parametric distributions (e.g. normal distribution or Student’s T-distribution). It has been proven that the ionospheric errors are of heavy-tailed nature and therefore are better described by a t-distribution than by a Gaussian distribution. Based on the results of the error analysis several variance-flooring strategies have been developed and tested. The variance-flooring strategies take into account the variances of the TEC-values as well as the variances of the service quality indicator QI and the variances of the mean values of QI of single receiver configurations. In addition, the 68th percentiles of the error distributions as well as the scales of the t-distribution have been used for the variance flooring. The developed variance-flooring strategies have been implemented into the receiver’s GNSS-firmware in the form of ionospheric constraint lookup tables. Large datasets of static as well as of dynamic test scenarios have been used to test the new variance-flooring strategies. The test scenarios covered high as well as low ionospheric activities. The variance-flooring strategies based on the 68th percentiles as well as on the t-distribution-scales clearly have shown an improved ambiguity resolution robustness as well as a decreased number of wrong fixes of the receiver. The positioning accuracy in 2D as well as in 3D is found to be comparable to the one of the basic receiver without any variance flooring. 2026-01-01T00:00:00Z http://trepo.tuni.fi:80/handle/10024/238910 A Digital Content Provenance Approach to Support Attribution, Copyright, Intellectual Property Claims, and Governance for Human-AI Created Works Mugabane, Baranaba Digital content is increasingly produced through workflows involving generative artificial intelligence, raising complex questions of authorship, copyright, disclosure, licensing, responsibility, and governance. These questions can only be answered meaningfully when reliable information exists about the people, tools, resources, and processes involved in the creation of a work. Yet a persistent informational gap remains. This research addresses that gap by designing and evaluating a governance-oriented provenance approach for human-AI-created works. The study adopts a Design Science Research Methodology and combines normative doctrinal analysis with structured artefact evaluation. Legal and regulatory analysis of the European Union, the United States, and the World Intellectual Property Organization framework is used to derive governance requirements, while existing provenance and traceability systems are reviewed to establish the technical landscape and inform the design. On that basis, the research develops a workflow-level provenance approach, including a new Entity-Action-Attribution data model and an extensible metadata structure for representing governance-relevant information such as AI involvement, rights reservations, and authorisation conditions. The approach is implemented and demonstrated in representative application scenarios, including a multimodal chatbot-based environment, and evaluated against the requirements derived from the legal and technical analysis. 2026-01-01T00:00:00Z http://trepo.tuni.fi:80/handle/10024/238868 Strictly Personalized Zero-Shot ECG Arrhythmia Detection Using Sparse Geometry-Regularized Autoencoders Haque, Asma Cardiovascular diseases remain the leading cause of mortality worldwide, and cardiac arrhythmias represent a major clinical concern due to their potential association with stroke, heart failure, and sudden cardiac death. While deep learning has achieved remarkable success in ECG arrhythmia classification, most existing approaches rely on large labelled datasets containing both normal and abnormal heartbeats. This assumption is often unrealistic in personalized monitoring scenarios, where a new user may provide only a short period of healthy ECG activity and no examples of future arrhythmias. This thesis investigates a strictly personalized zero-shot ECG anomaly detection framework in which only the first five minutes of healthy ECG data from each patient are used for model development. No abnormal beats are used during training, validation, threshold calibration, or model selection. The problem is formulated as reconstruction-based anomaly detection, where abnormalities are identified through deviations from a learned representation of patient-specific healthy cardiac activity. Three reconstruction-based models are evaluated under the same experimental protocol: a standard Autoencoder (AE), a Variational Autoencoder (VAE), and a proposed Sparse Geometry Autoencoder. The proposed model incorporates latent sparsity and geometry-preserving regularization to encourage compact and structured representations of healthy ECG morphology. A comprehensive ablation study investigates the effects of latent dimensionality, masking strategies, weight tying, sparsity regularization, geometry regularization, and threshold calibration. Experiments are conducted on the MIT-BIH Arrhythmia Database using patient-specific training and evaluation. Performance is assessed using accumulated confusion matrices and a macro-style F1-score to account for class imbalance. The proposed Sparse Geometry Autoencoder achieves the strongest overall performance, obtaining a macro-style F1-score of 0.8399, an accuracy of 0.9265, a precision of 0.7280, a recall of 0.7162, and a specificity of 0.9588. The results demonstrate that meaningful arrhythmia detection can be achieved using only a short healthy baseline and without requiring abnormal examples during model development. The findings indicate that compact latent representations combined with sparsity and geometry-preserving regularization improve reconstruction-based anomaly detection under strict personalized zero-shot constraints. More broadly, the study demonstrates the potential of patient-specific representation learning as a foundation for future personalized cardiac monitoring systems. 2026-01-01T00:00:00Z