Machine Learning and Stochastic Modelling Approaches for Biomedical Signal Analysis
This dissertation brings together a number of contributions on the application of Signal Theory and Data Analysis to solve clinical problems related to biomedical and biomechanical signals. The complexity of the processed signals, which is shaped by the underlying mechanisms of the biological process, conditions the nature of the proposed methods. Speci cally, the analysis of unpredictable signals such as activity patterns in certain conditions led to contributions in the machine learning field. These were applied to the identi cation of meaningful signal periods in activity registries and to support the diagnosis of the Attention De cit and Hyperactivity Disorder (ADHD), among other pathologies. On the other hand, this thesis proposes modelling approaches to deal with signals with well de ned evolution patterns (e.g., cardiac signals), so that higher applicability is achieved. With this regard, stochastic models and reconstruction methodologies based on them have been proposed enabling a wide range of applications in cardiology and other potential areas. The thesis is presented as a compendium of 12 peer-reviewed papers. Among them, 3 SCI JCR indexed journal papers comprise the core of the dissertation. The rest of them, have been published in the proceedings of top conferences in the biomedical engineering and/or signal processing elds, both at national (2 papers), and international level (7).