Browsing by Department "Department of Biomedical Engineering and Computational Science"

Filter results by typing the first few letters
Now showing 1 - 20 of 54
  • Advances in Approximate Bayesian Inference for Gaussian Process Models
    (2013) Riihimäki, Jaakko
     School of Science | Doctoral dissertation (article-based)
    Gaussian processes (GPs) provide a flexible approach to construct probabilistic models for Bayesian data analysis. In the Bayesian approach, GPs are used to specify prior assumptions on the latent function values that describe the underlying relationships between the explanatory variables and the associated target variables. These prior assumptions are combined with information from the observations using Bayes' rule. The obtained result is the posterior distribution that represents the uncertainty about the latent function values of interest, conditioned on the observations and the model assumptions. A challenge with the Bayesian approach is that exact inference is analytically intractable to calculate for most GP models used in practice. Therefore, approximate methods are needed in order to evaluate the posterior distribution and to make predictions for new observations. This thesis develops methods for approximate Bayesian inference in various modelling problems involving GP models. The focus is on efficient ways to form Gaussian posterior approximations based on Laplace's method or expectation propagation (EP). The inference for the studied GP models is challenging in two aspects. Firstly, observation models are generalized in the way that the probability distribution for each observation can depend on multiple values of the latent function instead of only one value, or on the derivative values of the latent function. Secondly, instead of one prior process, the models can have multiple uncorrelated prior processes that are coupled through the observation model. This thesis presents improvements to approximate Bayesian inference for GP models in density estimation, survival analysis, and multiclass classification. We describe Laplace's method for a logistic GP model and for a Cox-type survival model constructed from GP priors to speed up the inference. We develop a novel nested EP algorithm for multinomial probit GP classification that does not require numerical quadratures and scales linearly in the number of classes. In addition, we extend the existing methodology proposed for regression and binary classification by introducing monotonicity information into a GP model with EP. We demonstrate the practical accuracy of the described methods with several experiments and apply them to real-life modelling problems.
  • Approximate Bayesian Inference Methods for Regression and Classification with Gaussian Processes and Neural Networks
    (2013) Jylänki, Pasi
     School of Science | Doctoral dissertation (article-based)
    During the recent decades much research has been done on a very general approximate Bayesian inference framework known as expectation propagation (EP), which has been found to be a fast and very accurate method in many experimental comparisons. A challenge with the practical application of EP is that a numerically robust and computationally efficient implementation is not straightforward with many model specifications, and that there is no guarantee for the convergence of the standard EP algorithm. This thesis considers robust and efficient application of EP using Gaussian approximating families in three challenging inference problems. In addition, various experimental results are presented to compare the accuracy of EP with several alternative methods for approximate Bayesian inference. The first inference problem considers Gaussian process (GP) regression with the Student-t observation model, where standard EP may run into convergence problems, because the posterior distribution may contain multiple modes. This thesis illustrates the situations where standard EP fails to converge, reviews different modifications and alternative algorithms for improving the convergence, and presents a robust EP implementation that relies primarily on parallel EP updates and uses a provably convergent double-loop algorithm with adaptively selected step size in difficult cases. The second inference problem considers multi-class GP classification with the multinomial probit model, where a straightforward EP implementation requires either multi-dimensional numerical integrations or a factored posterior approximation for the latent values related to the different classes. This thesis describes a novel nested EP approach that does not require numerical integrations and approximates accurately all between-class posterior dependencies of the latent values, but still scales linearly in the number of classes. The third inference problem considers nonlinear regression using two-layer neural networks (NNs) with sparsity-promoting hierarchical priors on the inputs, where the challenge is to construct sufficiently accurate and computationally efficient approximations for the likelihood terms that depend in a non-linear manner on the network weights. This thesis describes a novel computationally efficient EP approach for simultaneous approximate integration over the posterior distribution of the weights, the hierarchical scale parameters of the priors, and the residual scale. The approach enables flexible definition of weight priors with different sparseness properties, and it can be extended beyond standard activation functions and NN model structures to form flexible nonlinear predictors from multiple sparse linear models.
  • Bayesian spatial and temporal epidemiology of non-communicable diseases and mortality
    (2011) Havulinna, Aki S
     Perustieteiden korkeakoulu | Doctoral dissertation (article-based)
    Spatial epidemiology combines spatial statistical modelling and disease epidemiology for studying geographic variation in mortality and morbidity. The effects of putative risk factors may be examined using ecological regression models. On the other hand, age-period-cohort models can be used to study the variation of mortality and morbidity through time. Bayesian hierarchical statistical models offer a flexible framework for these studies and enable the estimation of uncertainties in the results. The models are usually estimated using computer-intensive Markov chain Monte Carlo simulations. In this dissertation the first four publications present practical epidemiological studies on geographic variation in non-communicable diseases in Finland. In the last publication we study the long-time variation in all-cause mortality in several European countries. New statistical models are developed for these studies. This work provides new epidemiological information on the geographic variation of acute myocardial infarctions (AMI), ischaemic stroke and parkinsonism in Finland. An extended model for studying shared and disease specific geographic variation is presented using data on AMI and ischaemic stroke incidence. Existing results on the inverse association of water hardness and AMI are refined. New models for interpolation of geochemical data with non-detected values are presented with case studies using real data. Finally, the Bayesian age-period-cohort model is extended with versatile interactions and better prediction ability. The model is then used to study long-term variation in mortality in Europe.
  • Bioinformatics approaches for the analysis of lipidomics data
    (2010) Yetukuri, Laxmana Rao
     Informaatio- ja luonnontieteiden tiedekunta | Doctoral dissertation (article-based)
    The potential impact of lipid research has been increasingly realised both in disease treatment and prevention. Recent advances in soft ionization mass spectrometry (MS) such as electrospray ionization (ESI) have permitted parallel monitoring of several hundreds of lipids in a single experiment and thus facilitated lipidomics level studies. These advances, however, pose a greater challenge for bioinformaticians to handle massive amounts of information-rich MS data from modern analytical instruments in order to understand complex functions of lipids. The main aims of this thesis were to 1) develop bioinformatics approaches for lipid identification based on ultra performance liquid chromatography coupled to mass spectrometry (UPLC/MS) data, 2) predict the functional annotations for unidentified lipids, 3) understand the omics data in the context of pathways and 4) apply existing chemometric methods for exploratory data analysis as well as biomarker discovery. A bioinformatics strategy for the construction of lipid database for major classes of lipids is presented using simplified molecular input line entry system (SMILES) approach. The database was annotated with relevant information such as lipid names including short names, SMILES information, scores, molecular weight, monoisotopic mass, and isotope distribution. The database was tailored for UPLC/MS experiments by incorporating the information such as retention time range, adduct information and main fragments to screen for the potential lipids. This database information facilitated building experimental tandem mass spectrometry libraries for different biological tissues. Non-targeted metabolomics screening is often get plagued by the presence of unknown peaks and thus present an additional challenge for data interpretation. Multiple supervised classification methods were employed and compared for the functional prediction of class labels for unidentified lipids to facilitate exploratory analysis further as well as ease the identification process. As lipidomics goes beyond complete characterization of lipids, new strategies were developed to understand lipids in the context of pathways and thereby providing insights for the phenotype characterization. Chemometric methods such as principal component analysis (PCA) and partial least squares and discriminant analysis (PLS/DA) were utilised for exploratory analysis as well as biomarker discovery in the context of different disease phenotypes.
  • Brain Mechanisms Underlying Perception of Naturalistic Social Events
    (2014) Lahnakoski, Juha
     School of Science | Doctoral dissertation (article-based)
    Understanding the brain basis of the wide variety of skills needed to seamlessly interact with other people in the social world is one of the most important goals of social cognitive neuroscience. However, it has remained unclear how the elementary processes of social interaction that have been studied so far generalize to complex naturalistic settings where multiple social cues have to be dynamically tracked at the same time. The studies presented here employ movies to depict real-life-like social interactions and map the brain systems that participate in the perception of different aspects of the stimuli, particularly focusing on their social signals. Brain activity during viewing of the movies was recorded with functional magnetic resonance imaging. Additionally, the eye gaze of the subjects was recorded and behavioral measures were acquired in a subset of the experiments. Subjects' viewing patterns and interpretation of the movie were manipulated in one of the studies by asking the subjects to adopt two different perspectives. The brain activity was analyzed by multiple methods: parametric models of stimulus contents, inter-subject correlations of brain activity, independent component analysis, and network analysis. First study compared methods for analyzing both stimulus-related activity and covariation of networks of brain regions during naturalistic conditions to find networks responding to speech, sound, motion categories and low-level visual information. Second study elucidated the organization of the brain regions participating in processing several types of social contents. The results highlight the role of the posterior superior temporal sulcus as a key structure potentially integrating multiple types of social information. Finally, the third study provides first direct experimental evidence for the hypothesis that shared brain activity across individuals reflects shared understanding of the external world. Consistency of the findings demonstrate the feasibility of studying brain responses to simple stimulus features, social movie content as well as high-level perspective taking tasks during very rich naturalistic audiovisual stimulus conditions.
  • Cardiovascular magnetic resonance imaging: Error sources in phase contrast flow measurements
    (2013) Peltonen, Juha
     School of Science | Licentiate thesis
    Phase contrast velocity encoded flow measurement was first introduced 30 years ago and have been subject to constant evolution since. The method has been validated repeatedly to be suitable for numerous clinical applications. It is capable to visualize and quantify dynamic phenomena such as blood flow in any part of the human body without being invasive or using ionizing radiation. Recent publications even suggest that velocity encoded imaging should be used as a gold standard in the flow measurements of great arteries. Despite of numerous possibilities, the method remains underused. However, phase contrast flow imaging carries some challenges that are limiting clinical applicability. First of all, magnetic resonance imaging is expensive and time consuming imaging modality. Currently velocity encoded flow imaging can be seen as a secondary modality which is used if necessary after alternative methods such as ultrasound. Also, phase contrast flow imaging includes numerous physical and physiological error sources affecting the accuracy of flow measurements. The understanding of these error sources is important to be able to estimate the accuracy of obtained results. An overview on these error sources is presented in this work. Essential theoretical basis to understand the physics underlying the error mechanisms are presented and means to minimize the error are addressed. The dimensions of human arteries are varying considerably along the flow track beyond the ventricles. Changing geometrical dimension and individual characteristics of cardiac valves are giving a raise to several spatially alternating flow phenomena such as acceleration artefact and voxel dephasing. In this work a research is presented where optimal measurement plane to quantify stroke volume is studied. The importance of optimal measurement plane is studied in both healthy controls and patients with accelerated flow velocities. In the work it was found that optimal location to measure arterial flow is approximately 2 cm distal from the aortic or pulmonal valve in case of accelerated flow. In controls no such relation was found between the measurement plane and the stroke volume. We are also considering the amount of hardware velocity offset in our measurement system and discussing how it affects to the results.
  • Combined ultra-low-field MRI and MEG: instrumentation and applications
    (2013) Vesanen, Panu
     School of Science | Doctoral dissertation (article-based)
    Magnetic resonance imaging (MRI) is a noninvasive method that allows the study of the interior structure of matter. Today, MRI is widely used in medical diagnosis and research, thanks to its versatile contrast and the lack of ionizing radiation. Conventionally, the signal-to-noise ratio of an MRI measurement scales with the strength of the applied magnetic field. This has driven the development of MRI scanners towards fields of 3 T and above. Ultra-low-field (ULF) MRI is an emerging technology that uses microtesla-range magnetic fields for image formation. The low signal-to-noise ratio is partly compensated for by prepolarizing the sample in a field of 1 – 200 mT and using superconducting quantum interference devices (SQUIDs) for signal detection. Advantages of ULF MRI include unique low-field contrast mechanisms, flexibility in the sequence design, and the possibility to construct a silent scanner with an open geometry. ULF MRI is also compatible with magnetoencephalography (MEG), which uses SQUIDs to record the magnetic field produced by neuronal activity. With a hybrid scanner combining MEG and MRI, both the structure and function of the human brain can be studied with a single device. In this Thesis, a hybrid MEG-MRI device was designed, constructed, and tested. The system is based on a commercial whole-head MEG device that was modified to accommodate ULF-MRI functionality. In particular, the effects of the various magnetic fields applied inside a magnetically shielded room were studied. To prevent the harmful effects of the eddy currents caused by changing magnetic fields, a self-shielded polarizing coil was designed and constructed. Moreover, the conventional SQUID design was modified in order to develop sensor modules that tolerate the relatively strong polarizing field. Finally, the device was used to measure MEG data and ULF-MR images of the human brain. In addition to the instrumentation development, several applications of ULF MRI were investigated. A method for imaging electric current density was presented. The technique takes advantage of the flexibility of ULF MRI by encoding the signal in zero magnetic field. Furthermore, the temperature dependence of the MRI relaxation times was studied. Drastic variations were found as a function of the field strength. The results were used to reconstruct temperature maps using ULF MRI. The results presented in this Thesis demonstrate that upgrading MRI functionality into an existing commercial MEG device is a feasible concept. Such a device has the potential to enable new methods and paradigms for neuroscientific research. The possibility of taking advantage of the unique low-field contrast is an interesting subject for further research.
  • Computational analysis of large and time-dependent social networks
    (2013) Kovanen, Lauri
     School of Science | Doctoral dissertation (article-based)
    Complex systems consist of a large number of elements that interact in a non-trivial way; for example the human brain, society, Internet, and biological organisms can all be modelled as complex systems. Complex systems can be naturally represented as networks, mathematical objects that consist of nodes and edges connecting these nodes, and the study of large networks based on empirical data has become known as complex networks. Since the first articles on complex networks appeared in the end of the 1990's, various technological, biological, and social networks have been analyzed. In recent years introductory text books on the subject have also been published.  The study of social networks of course has a longer history. Small social networks have been studied for decades in sociology, social psychology and anthropology, and the influence that social networks have on both performance and well being of individuals has been well documented. The availability of electronic communication records—mobile phone calls, emails, online social networking sites and even multiplayer computer games—have changed the scale and detail at which social networks can be analyzed. The largest data set studied so far includes over 700 million individuals, and the mobile phone call records studied in this Thesis contain information of over 6 million people. The combination of powerful computers and large data sets have enabled the emergence of computational social science. Several aspects of large social networks are studied in this Thesis. Models of social networks are commonly used as a way to gain insight about the structure of these networks. The first article studies a number of models suggested for social networks and discusses their advantages and shortcomings. The community structure of various networks has also been a subject of great interest. It is widely accepted that nearly all networks have modular structure, evidenced by local densifications of connectivity. However, identifying communities in empirical data has turned out to be difficult both theoretically and in practice. We apply three state-of-art community detections methods to a large social network and evaluate the quality of the identified communities. One important aspect of human interactions is omitted when analyzing networks: time. Temporal networks have become a common framework for studying data sets where the relations between nodes vary with time, and this framework can be readily applied to study mobile phone calls. The last part of this Thesis introduces the concept of temporal motifs—recurring patterns of events in temporal networks—that can be used to analyze the meso-scale structure of temporal networks.
  • Computational analysis of the metabolic phenotypes in type 1 diabetes and their associations with mortality and diabetic complications
    (2010) Mäkinen, Ville-Petteri
     Aalto-yliopiston teknillinen korkeakoulu | Doctoral dissertation (article-based)
    Type 1 diabetes is an autoimmune disease that destroys the secretion of insulin (in the pancreas); insulin is a vital hormone for maintaining normal glucose metabolism. Insulin replacement therapy can prevent the acute symptoms, but is not able to fully match the natural regulation, which puts a metabolic stress on tissues. For some patients, the stress manifests as gradual damage to blood vessels and the nervous system over the next few decades after diabetes diagnosis. The aim of the thesis was to describe the metabolic profiles and to investigate their connections with the spectrum of clinical symptoms. Simultaneously, new techniques were applied to measure the profiles (1H NMR spectroscopy) and to visualize the multivariate statistical associations (the self-organizing map). A total of 4,197 patients with type 1 diabetes were recruited for the thesis by the Finnish Diabetic Nephropathy Study. A quarter of the patients exhibited an obesity-related phenotype (high triglycerides, cholesterol, apolipoprotein B-100, low high-density lipoprotein cholesterol, high C-reactive protein). A third of the individuals had a diabetic kidney disease phenotype (high urinary albumin and serum creatinine). The combination of the two was associated with a 10-fold population-adjusted mortality. Nevertheless, there was no discernible metabolic threshold between the phenotype models, nor were there any single variable that could predict the outcomes accurately. These results suggest a need for multifactorial and multidisciplinary paradigms for the research, treatment and prevention of diabetic complications.
  • Computational models and methods for lipoprotein research
    (2011) Kumpula, Linda
     Perustieteiden korkeakoulu | Doctoral dissertation (monograph)
    Lipoproteins are self-assembled nanoparticles for water-insoluble lipid transportation in the circulation. Lipoprotein particles form a key metabolic system in a variety of normal physiological processes but also play an essential role in many pathological conditions. In particular, certain lipoprotein abnormalities are associated with the development of atherosclerosis, a disease state of arteries, common in cardiovascular disease. Computational modelling is a potential but so far rarely used method to study lipoprotein particles. This thesis contributes to lipoprotein research by various computational approaches where experimentally isolated and biochemically characterised lipoprotein particles serve as a starting point. This thesis deals with estimating the number of lipid molecules within lipoprotein particles, i.e., composition information, and approximating the molecular structure of lipoprotein particles in each subclass. It also proceed the ultracentrifugal particle isolation by a kind of in silico sub-classification resulting from utilisation of the self-organising map (SOM) method. This, when applied to experimental data, with lipoprotein lipid concentration and composition information combined, shows that there is variability in the compositional/metabolic relations between individuals, i.e., distinct lipoprotein phenotypes. Furthermore, this thesis introduces a method to estimate lipoprotein particle concentrations in each subclass, which also provides a reference particle library for NMR-based lipoprotein particle concentration estimation. Applications of the models to experimental data show that triglyceride and cholesterol ester molecules, which are conventionally held as core lipids, may also locate in significant amounts in the surface. The lipoprotein phenotype analysis shows that per particle compositions, which appear as a fundamental issue in metabolic and clinical corollaries, can not be deduced solely from the regularly measured plasma lipid concentrations nor from the particle concentration estimates.
  • Control of Protein Oligomerization and De-oligomerization on Lipid Membranes
    (2013) Mahalka, Ajay
     School of Science | Doctoral dissertation (article-based)
    Oligomerization of protein into amyloid fibrils is central to the pathogenesis of several neurodegenerative disorders. Amyloid fibrillation and the cytotoxic actions of amyloids are membrane-associated processes. The interactions of amyloid-forming proteins with lipids at the membrane surface accelerate fibrillation and induce membrane permeabilization. Oligomerization also plays a functional role in antimicrobial defense and controls the catalytic activity of phospholipase A2 (PLA2). The protein oligomerization and amyloid formation can be modulated by heat shock protein 70 (Hsp70). Thus, the aim of the present work was to study membrane-associated protein oligomerization and its modulation by Hsp70 on the phospholipid model membrane system. Sequence analyses revealed that antimicrobial peptides (AMPs) contained sequence motifs that showed propensities for self-assembly, aggregation, and oligomerization into amyloid fibrils. The presence of such oligomerization-mediating sequences was characteristic of amyloidogenic cytotoxic proteins, including gelsolin involved in familial Finnish type amyloidosis (FAF). 1-Palmitoyl-2-(9'-oxo-nonanoyl)-sn-glycero-3-phosphocholine (PoxnoPC), an oxidized phospholipid, accelerated fibrillation of the core amyloidogenic segment of gelsolin.The PoxnoPC-mediated fibrillation of gelsolin was dependent on both the concentration and the aggregation state of PoxnoPC. Fibril growth followed simple nucleation-dependent kinetics with the formation of transient prefibrillar oligomers in the lag phase. Subsequently, in order to understand the functional role of membrane-associated Hsp70, we studied lipid-Hsp70 interactions. The association of Hsp70 with phospholipid membranes was highly dependent on their lipid compositions. Hsp70 associated with phosphatidylcholine bilayers and penetrated into the hydrocarbon region. In contrast to the above data, in the presence of negatively charged phospholipids, Hsp70 bound peripherally to membrane surfacesby extended phospholipid anchorage. A specific pH-dependent association of Hsp70 with bis(monoacylglycero)phosphate, an acidic phospholipid enriched in the inner lysosomal membrane, activated lysosomal acid sphingomyelinase and promoted cell survival. We also showed that the Hsp70 sustained the hydrolytic activity of PLA2 by modulating the oligomerization and transformation of PLA2 into amyloid fibers. Hsp70 attenuated the lysophosphatidylcholine-induced inhibition and amyloid formation of PLA2 in an ATP-dependent manner. Finally, an oligomerization-mediating sequence in PLA2 was identified. Synthetic peptides corresponding to amyloidogenic, aggregation-promoting regions inhibited the hydrolytic activity of PLA2.
  • Data integration, pathway analysis and mining for systems biology
    (2010) Peddinti, Venkata Gopalacharyulu
     Informaatio- ja luonnontieteiden tiedekunta | Doctoral dissertation (article-based)
    Post-genomic molecular biology embodies high-throughput experimental techniques and hence is a data-rich field. The goal of this thesis is to develop bioinformatics methods to utilise publicly available data in order to produce knowledge and to aid mining of newly generated data. As an example of knowledge or hypothesis generation, consider function prediction of biological molecules. Assignment of protein function is a non-trivial task owing to the fact that the same protein may be involved in different biological processes, depending on the state of the biological system and protein localisation. The function of a gene or a gene product may be provided as a textual description in a gene or protein annotation database. Such textual descriptions lack in providing the contextual meaning of the gene function. Therefore, we need ways to represent the meaning in a formal way. Here we apply data integration approach to provide rich representation that enables context-sensitive mining of biological data in terms of integrated networks and conceptual spaces. Context-sensitive gene function annotation follows naturally from this framework, as a particular application. Next, knowledge that is already publicly available can be used to aid mining of new experimental data. We developed an integrative bioinformatics method that utilises publicly available knowledge of protein-protein interactions, metabolic networks and transcriptional regulatory networks to analyse transcriptomics data and predict altered biological processes. We applied this method to a study of dynamic response of Saccharomyces cerevisiae to oxidative stress. The application of our method revealed dynamically altered biological functions in response to oxidative stress, which were validated by comprehensive in vivo metabolomics experiments. The results provided in this thesis indicate that integration of heterogeneous biological data facilitates advanced mining of the data. The methods can be applied for gaining insight into functions of genes, gene products and other molecules, as well as for offering functional interpretation to transcriptomics and metabolomics experiments.
  • Data-analysis perspectives on naturalistic stimulation in functional magnetic resonance imaging
    (2010) Malinen, Sanna
     Aalto-yliopiston teknillinen korkeakoulu | Doctoral dissertation (article-based)
    Modern brain imaging allows to study human brain function during naturalistic stimulus conditions, which entail specific challenges for the analysis of the brain signals. The conventional analysis of data obtained by functional magnetic resonance imaging (fMRI) is based on user-specified models of the temporal behavior of the signals (general linear model, GLM). Alongside these approaches, data-based methods can be applied to model the signal behavior either on the basis of the measured data, as in seed-point correlations or inter-subject correlations (ISC), or alternatively the temporal behavior is not modeled, but spatial signal sources and related time courses are estimated directly from the measured data (independent component analysis, ICA). In this Thesis, fMRI data-analysis methods were studied and compared in experiments that gradually proceeded towards more naturalistic and complex stimuli. ICA showed superior performance compared with GLM-based method in the analysis of naturalistic situations. The particular strengths of the ICA were its capability to reveal activations when signal behavior deviated from an expected model, and to show similarities between signals of different brain areas and of different individuals. The practical difficulty of ICA in naturalistic conditions is that the user may not be able to determine, purely on the basis of the components' spatial distribution or temporal behavior, the brain networks that are related to the given stimuli. In this Thesis, a new solution to sort the components was proposed that ordered the components according to the ISC map, and thereby facilitated the selection of stimulus-related components. The method prioritized brain areas closely related to sensory processing, but it also revealed circuitries of intrinsic processing if they were affected similarly across individuals by external stimulation. Analysis issues related to the impact of physiological noise in fMRI signals were also considered. Cardiac-triggered fMRI improved detection of touch-related activation both in the thalamus and in the secondary somatosensory cortex. The most common way to eliminate noise is to filter the data. In this Thesis, however, aberrations in temporal behavior, as well as in functional connectivities in chronic pain patients were observed, which likely could not have been revealed with conventional temporal filtering.
  • Developing neurophysiological metrics for the assessment of mental workload and the functional state of the brain
    (2010) Holm, Anu
     Aalto-yliopiston teknillinen korkeakoulu | Doctoral dissertation (article-based)
    Modern working environments are often information intensive and work performance requires acting on multiple tasks simultaneously, i.e., multitasking. Also, irregular and prolonged work schedules, shift work and night work are typical in many work sectors. This causes both acute and chronic sleep loss, which results in performance impairment, such as increased reaction times, memory difficulties, cognitive slowing, and lapses of attention. Long lasting sleep loss and sustained overloading increase the risk of human errors and may cause work related stress and even occupational burn-out. According to the Finnish Occupational Safety and Health Act (738/2002, Työturvallisuuslaki), Section 25 Avoiding and reducing workloads, an employer should assess the workload the employee is exposed to. Despite the fact that this important issue is enacted in the law, the objective measures to assess the workload are lacking. This Thesis reviews neurophysiologic methods for assessment of cognitive workload and sleep loss. Then it describes experimental studies where the feasibility of conventional event related potential (ERP) and electroencephalography (EEG) methods were tested both in assessment of internal state of participants during challenging task performance after sleep debt and in diagnostic of work-related central nervous system disorder. After that, methodological improvements both on ERPs and EEG metrics are shown: ERPs were analysed with a single-trial method, and EEG methodology was developed for estimation of both internal (caused by sleep loss) and external (caused by task demands) load. The methods were tested in healthy controls. The most promising metric to study overall brain load, including both cognitive workload as well as sleep loss, is suggested to be theta Fz / alpha Pz -ratio. It increases both with growing cognitive workload level and time spent awake, being sensitive also to sleep loss. This metric is possible to measure both in laboratory and in the field conditions. Measurements may be carried out even during real work tasks, at least in professions where most work is done in office-like environments. As the ratio increases with cognitive brain load similar to the heartbeat with increasing physical load, the ratio was named "brainbeat".
  • Developmental biology of sex determination : establishing a basis for systems approach
    (2008-03-14) Segerståhl, Margareta
     Faculty of Information and Natural Sciences | Doctoral dissertation (monograph)
    The existence of males and females is a fundamental aspect of animal biology but the developmental mechanisms of sex determination and sexual differentiation exhibit surprising evolutionary diversity. Depending on the species, sex determination mechanisms range from genetic sex chromosome effects to environmental temperature effects and even to social cues that derive from local population structure. Many genes and molecules that play a role in sex determination biology have been identified, yet it has often remained unclear how these factors work together. In this thesis, developmental biology of sex determination is approached as a complex dynamic system. The scope is widened from animal biology to an examination of a general biological phenomenon, complexity of which is addressed directly. The method of investigation is a theoretical one relying mostly on heuristic problem solving methods and system identification procedures. Posing of novel questions enables formulation of alternative hypothesis about some fundamental aspects of sex determination biology. For instance, according to current view, two distinctly different developmental programs form ovaries and testes. Here, an integrating system description is provided where both testicular and ovarian development are derivatives of a common developmental scheme. Also, in this thesis sexual germ line development receives a much more prominent role than what is usually the case in sex determination studies. A premis is formed that the foundation of system properties of sex determination biology can be found in the developmental life cycle progression of the germ cell line. This enables formulation of testable prediction about the nature of the relationship between germ cell biology and developmental mechanisms of complex multicellularity. This work shows that a dynamical systems view can be applied to developmental biology and sex determination studies. Conceptual tools are provided for the attainment of a better understanding of genetic and molecular aspects of sex determination biology and sexually dimorphic development.
  • Dynamic correlations in ongoing neuronal oscillations in humans - perspectives on brain function and its disorders
    (2010) Monto, Simo
     Aalto-yliopiston teknillinen korkeakoulu | Doctoral dissertation (article-based)
    This Thesis is involved with neuronal oscillations in the human brain and their coordination across time, space and frequency. The aim of the Thesis was to quantify correlations in neuronal oscillations over these dimensions, and to elucidate their significance in cognitive processing and brain disorders. Magnetoencephalographic (MEG) recordings of major depression patients revealed that long-range temporal correlations (LRTC) were decreased, compared to control subjects, in the 5 Hz oscillations in a manner that was dependent on the degree of the disorder. While studying epileptic patients, on the other hand, it was found that the LRTC in neuronal oscillations recorded intracranially with electroencephalography (EEG) were strengthened in the seizure initiation region. A novel approach to map spatial correlations between cortical regions was developed. The method is based on parcellating the cortex to patches and estimating phase synchrony between all patches. Mapping synchrony from inverse-modelled MEG / EEG data revealed wide-spread phase synchronization during a visual working memory task. Furthermore, the network architectures of task-related synchrony were found to be segregated over frequency. Cross-frequency interactions were investigated with analyses of nested brain activity in data recorded with full-bandwidth EEG during a somatosensory detection task. According to these data, the phase of ongoing infra-slow fluctuations (ISF), which were discovered in the frequency band of 0.01-0.1 Hz, was correlated with the amplitude of faster > 1 Hz neuronal oscillations. Strikingly, the behavioral detection performance displayed similar dependency on the ISFs as the > 1 Hz neuronal oscillations. The studies composing this Thesis showed that correlations in neuronal oscillations are functionally related to brain disorders and cognitive processing. Such correlations are suggested to reveal the coordination of neuronal oscillations across time, space and frequency. The results contribute to system-level understanding of brain function.
  • Dynamic tactile feedback in human computer interaction
    (2012) Ahmaniemi, Teemu
     School of Science | Doctoral dissertation (article-based)
    Vibrotactile stimulation is an efficient way to provide feedback when a user is interacting with a mobile device. Because the device is held in hand, the user's input, such as touch and gesturing, can be easily measured with force- and motion sensors. In addition, the feedback can be provided into the same device with a vibrotactile actuator. This thesis introduces a set of methods to provide the vibrotactile feedback dynamically i.e. proportionally to the applied input in real-time during the interaction. The effects of dynamic tactile feedback in gesture and touch interaction are investigated in five studies reported in the thesis. The first two studies introduce the method for providing the dynamic feedback based on wavetable synthesis and assess the perceptual dimensions of the feedback when coupled with gesture movements. The studies indicated that the method is a powerful tool for creating an illusion of real textures with a simple vibrotactile actuator and motion sensor. Modifications in texture ridge length and spatial density lead into large variation in subjective evaluations of texture roughness, bumpiness, stickiness and pleasantness. In study III, mobile device was used to search for targets in the environment. Tactile feedback turned out to be an efficient way to indicate target directions. The finding hints that vibration could be efficiently used as a private and transparent signal to indicate directions and target locations in augmented reality interaction. In study IV, dynamic tactile feedback was provided when playing a gesture controlled virtual musical instrument. The feedback reflecting the distance to a virtual control surface assisted the temporal accuracy of the playing. In study V, vibrotactile feedback was provided proportionally to the applied force on a rigid surface. In a force repetition task, all the dynamic feedback models assisted the accuracy of the force control. On contrast, in a force hold task, the feedback did not assist the performance. Dynamic tactile feedback appeared to be a powerful tool to create illusions of real textures and it improved the motor accuracy in repetitive gesture- and force based interaction. The findings make a contribution to research focusing on touch and gesture based user interfaces as well as psychophysics research dealing with augmented feedback. In order to apply the findings in future user interfaces of mobile devices, more studies are needed. In further work, system latency should be minimized and tactile feedback should be design in parallel with visual and auditory feedback.
  • Dynamics of single biopolymer translocation and sedimentation
    (2010) Lehtola, Ville
     Aalto-yliopiston teknillinen korkeakoulu | Doctoral dissertation (article-based)
    In this Thesis the dynamics of translocation and sedimentation of a single biopolymer (typically DNA, RNA or a protein) is studied. A coarse-graining paradigm is invoked to justify the various computational models by use of which the results are obtained. The transport of biopolymers through a nano-scale pore in a membrane is a ubiquitous process in biology. Experimental interest in translocation process focuses on its potential applicability in ultra-fast sequencing of DNA and RNA molecules. Polymer translocation has been under intense study for over a decade. Inspite of the vast theoretical research, the experimental results on the driven case have not been explained. We claim that the reason for this is that the translocation process must be treated as (at least) two dynamically distinct cases, where the dynamics takes place either close to or out of equilibrium. Here, we find that the latter case corresponds to the experiments. We make a comprehensive investigation on how the process can be discriminated based on its dynamics, and define and use some indicators to this end. In addition, we study the role of hydrodynamics, and find it to govern the dynamics when the process takes place out of equilibrium. Sedimentation is a natural process induced by gravity that can be applied experimentally in a quickened form by the use of ultracentrifuges, and which is similar to electrophoresis. Our study on behavior of single polymers under non-equilibrium conditions falls within the rapidly developing field of nano- and microfluidics that has important applications in "lab-on-a-chip" based technologies. In polymer sedimentation, we study the settling of the polymer in a steady-state in the limit of zero Péclet number, i.e. where no thermal fluctuations exist. The hydrodynamic coupling of the polymer beads leads to chaotic time-dependent behavior of the chain conformations that in turn are coupled with the velocity fluctuations of the polymer's center of mass.
  • Event detection in preterm electroencephalography
    (2014) Palmu, Kirsi
     School of Science | Doctoral dissertation (article-based)
    Preterm infants may spend months in neonatal intensive care units (NICU). Progress in neurological care of these infants depends on the ability to adequately monitor brain activity during NICU treatment. Brain monitoring is most commonly performed using electroencephalography (EEG). The preterm EEG signals are qualitatively different from EEG signals of older individuals, their distinguishing characteristics are the intermittently occurring spontaneous activity transients (SAT), which are believed to be crucial to early brain development. Automated detection of SATs might offer new tools for a neuroscientifically reasoned monitoring of infant brain in the NICU.  In this Thesis, a commercially available algorithm was tested for its applicability in detecting SATs. Because the algorithm was found to be suboptimal, an improved algorithm was developed and its parameters were optimized. Optimization and validation were done systematically, using a gold standard composed of unanimous detections by three human raters. The optimized algorithm was then used to calculate event-based measures in two clinical studies, one studying SAT occurrence in sleep stages, and the other comparing brain activity to structural brain growth.  In leave-one-out crossvalidation, the optimized algorithm showed excellent performance (sensitivity 96.6±2.8 %, specificity 95.1±5.6 %). In the clinical studies conducted, the proportion of EEG covered by SATs (SAT%) was shown to differ between sleep states, providing a possibility for developing an EEG-based measure of brain activity cycling in preterm infants. Finally, brain activity indices derived from EEG recordings shortly after birth were shown to correlate with subsequent structural growth of the brain during preterm life. The findings together show that an SAT event detector can be constructed for the brain monitoring in NICU, and that indices based on event detection may offer important insight to brain function in the clinical research.
  • Eye tracking Based Methods for Evaluation of Infants’ Visual Processing
    (2015) Ahtola, Eero
     School of Science | Licentiate thesis
    Cortical visual processing and mechanism under eye movements and visiospatial attention undergo prominent developmental changes during the first 12 months of infancy. At that time, these key functions of vision are tightly connected to the early brain development in general. Thus, they are favourable targets for new research methods that can be used in treatment, prediction, or detection of various adverse visual of neurocognitive conditions. This thesis presents two eye tracker assisted test paradigms that may be used to evaluate and quantify different functions of infants’ visual processing. The first study concentrates on the analysis of the gaze patterns in classic face-distractor competition paradigm known to tap mechanisms under infant’s attention disengagement and visuospatial orienting. A novel stimuli over a given period of time. In further evaluation, the metric is shown to be sensitive to developmental changes in infants’ face processing between 5 and 7 months of age. The second study focuses on the visual evoked potentials (VEPs) elicited by orientation reversal, global form, and clobal motion stimulation known to measure distinct aspects of visual processing at the cortical level. To improve the reality of such methods, an eye tracker is integrated to the recording setup, which can be used to control stimulus presentation to capture the attention of the infant, and in the analysis to exclude the electroencephalography (EEG) segments with disorientated gaze. With this setup, VEPs can be detected from the vast majority of the tested 3-month-old infants (N=39) using circular variant of Hotelling’s T2 test statistic and two developed power spectrum based metrics. After further development already in progress, the presented methods are ready to be used clinically in assessments of neurocognitive development, preferably alongside other similar biomarker tests of infancy.