Aktuální semináře ISCB ČR

Visualisation of spatially smoothed biological signal recorded on landmarks, curves or surfaces

přidáno: 1. 10. 2018 5:35, autor: Zdeněk Valenta   [ aktualizováno 8. 11. 2018 3:53 ]

Stanislav Katina, Zdeňka Geršlová, Vojtěch Šindlář, Mojmír Vinkler
Ústav matematiky a statistiky
Přírodovědecká fakulta 
Masarykova univerzita Brno
Kotlářská 267/2
611 37 Brno

Spatial interpolation and smoothing is usually done for one surface. In our case, we have random samples of such surfaces from three different domains, i.e. electrical activity of human brain and heart, and human faces. First domain, electrical activity of the human brain, characterised by event-related potentials (ERPs) can be recorded by electroencephalography (EEG) using scalp electrodes mounted on an elastic cap where the 61 electrode sites formed an equidistant grid. For the purpose of visualisation, this number of points is quite low and therefore we use 100,000 points. Second domain, electrical activity of the human heart, characterised by body surface potential maps (BSPMs) can be recorded by electrocardiography (ECG) using electrodes mounted on the skin around the chest where the 80 electrode sites formed a regular grid of 5 rows and 16 columns. For the purpose of visualisation, this number of points is quite low and therefore we use 83,520 points (145 × 576 grid). Third domain, human faces, captured by stereo-photogrammetry are characterised by about 150,000 points. These points are triangulated by about 300,000 triangles. This number of points is extremely high for the purpose of statistical analyses, therefore the coordinates of landmarks and curves sufficiently characterising the shape have to be automatically identified and this simplified model comprising about 1000 points is then used in further statistical modelling.

In this presentation we define the model for creating a grid of points in 3D for one subject that allows us to smooth the biological signal between the (semi)landmarks (coordinates of sensors, i.e. EEG/ECG electrodes, coordinates of biological (semi)landmarks) and then interpolate the signal using thin-plate splines. Then we visualise the brain activity on a sphere (simplification of the head) and consider the projection from sphere to the Montreal Neurological Institute (MNI) template of the brain, then the ECG signal on cylinder (simplification of the torso), and human face. We also visualise directional derivatives and the shape index, the measures of local surface topology. Finally, with the methods mentioned above, we create static images of the EEG/ECG signal and human faces in several different views, i.e. frontal, lateral, vertical, oblique or other, and animate this signal across chosen directions and views in the statistical software R. These images allow for an overview of a brain/heart activity or a shape of human face and ease the bio-medical interpretations. 

Keywords: spline smoothing, interpolation, warping, shape index, visualisation, EEG, ECG, human face

PřednášejícíDoc. PaedDr. RNDr. Stanislav Katina, Ph.D.Ústav matematiky a statistiky, Přírodovědecká fakulta Masarykovy univerzity v Brně, Kotlářská 267/2, 611 37 Brno

Datum: Čtvrtek 8. listopadu 2018

Čas: 13:00 hod

A time-domain-constrained fuzzy clustering method in biomedical data analysis

přidáno: 25. 9. 2018 4:45, autor: Zdeněk Valenta   [ aktualizováno 25. 9. 2018 9:29 ]

Aleksander Owczarek, Ph.D., DSc., Eng.
Head of the Department of Statistics
Department of Instrumental Analysis
Faculty of Pharmacy and Laboratory Medicine in Sosnowiec
Medical University of Silesia in Katowice

AbstractThe lecture deals with a presentation of a fuzzy clustering method with time-domain-constraints used in biomedical data analysis. Proposed method makes it possible to include natural constraints for biomedical data (especially biomedical signals) analysis using fuzzy clustering, that is, the neighboring samples of data belong to the same cluster. The method introduces two approaches to include the above kind of constraints. The first approach leads to the time-domain-constrained fuzzy c-regression models method. The second approach leads to the epsilon-insensitive version of the above method, which results in additional robustness for outliers and non-Gaussian noise. Finally, simulations on synthetic as well as real-life data and signals would be used to evaluate the performance and to show usefulness of the time-domain-constrained fuzzy clustering methods..  

Keywords: fuzzy clustering, switching regression models, biomedical data analysis

PřednášejícíAleksander Owczarek, Ph.D., DSc., Eng.
Místo: Ústav informatiky AV ČR, v.v.i., Pod Vodárenskou věží 2, 182 07 Praha 8, místnost č. 222
Datum: Čtvrtek 1. listopadu 2018
Čas: 14:00 hod

Semináři bude předcházet Valná hromada členů ISCB ČR, která proběhne od 13:30 hod na témže místě. Na programu bude volba předsedy, místopředsedy, hospodáře a revizora společnosti.

Identifying influential observations in joint Bayesian models

přidáno: 2. 3. 2018 5:54, autor: Zdeněk Valenta   [ aktualizováno 5. 4. 2018 2:02 ]

Mgr. Šárka Rusá, Dept. of Probability and Mathematical Statistics, Faculty of Mathematics and Physics, Charles University in Prague, Sokolovská 83, Praha 8-Libeň, Czech Republic

AbstractAlthough increasingly complex Bayesian models are being employed in practice, most of existing literature has not dealt with model diagnostics. We propose a Bayesian approach to the detection of influential observations or sets of observations in joint models. 
    Moreover, the usage of our method is advantageous if the investigation of case-deletion is restricted to a subset of parameters. Importance sampling with weights which take advantage of the dependence structure in hierarchical models is utilised in order to estimate the case influence on the parameters. 
    The method is illustrated on a three-level dataset from the field of nursing research which was previously used to fit a mediation model of patient satisfaction with care. We focused on case-deletion on both the second and the third level of the data.  

Keywords: Bayesian joint models; Influential observations, Importance sampling.

Přednášející: Mgr. Šárka Rusá, KPMS, MFF UK Praha, Sokolovská 83
Místo: Ústav informatiky AV ČR, v.v.i., Pod Vodárenskou věží 2, 182 07 Praha 8, místnost č. 222
Datum: Čtvrtek 5. dubna 2018
Čas: 13:30 hod

Semináři bude předcházet Valná hromada členů ISCB ČR, která proběhne od 13:00 hod na témže místě. Na programu bude zpráva o činnosti a hospodaření společnosti za rok 2017. 

Measurement theory and statistical models

přidáno: 28. 8. 2017 13:12, autor: Zdeněk Valenta   [ aktualizováno 28. 8. 2017 13:22 ]

Zhiliang Ying, Department of Statistics, Columbia University, 1255 Amsterdam Avenue, Room 1033 , School of Social Work Building, New York, NY 10027, U.S.A.

AbstractMeasurement theory has played a foundational role in educational, psychological and psychiatric assessment.This talk will introduce various statistical models that have been serving as key tools in measurement theory. It will provide detailed discussions on existing and new statistical models and statistical inferences thereof, with special focus on certain latent class and latent factor models as well as their extensions for categorical and counting process data. The new developments will be applied to examples in educational assessment and psychological evaluation.  

Keywords: Measurement theory; latent class model; categorical data; counting processes; psychometrics.

PřednášejícíProf. Zhiliang Ying, Ph.D., Professor of Statistics, Director of Graduate Studies, Department of Statistics, Chief Co-Editor of Statistica Sinica
Místo: Praktikum KPMS, MFF UK Praha, Sokolovská 83, 1. patro
Datum: Čtvrtek 5. října 2017
Čas: 15:40 hod

Teaching psychometrics and analyzing composite measurements with R and Shiny

přidáno: 19. 6. 2017 2:42, autor: Zdeněk Valenta   [ aktualizováno 19. 6. 2017 4:35 ]

Patrícia Martinková [1], Adéla Drabinová [1,2], Jakub Houdek [3], Lubomír Štěpánek [3,4]

[1] Oddělení medicínské informatiky a biostatistiky, Ústav informatiky AV ČR v.v.i., Pod Vodárenskou věží 2, 182 07 Praha 8
[2] Katedra pravděpodobnosti a matematické statistiky, Matematicko-fyzikální fakulta, Univerzita Karlova, Sokolovská 83, 186 75 Praha 8
[3] Fakulta informatiky a statistiky, Vysoká škola ekonomická v Praze, Nám W. Churchilla 4, 130 67 Praha 3
[4] 1. lékařská fakulta, Univerzita Karlova, Kateřinská 32, 121 08 Praha 2

AbstractThis work introduces ShinyItemAnalysis (Martinková et al. 2017) R package and an online shiny application for psychometric analysis of educational tests and their items, and difNLR (Drabinová et al., 2017) R package for detection of differential item functioning (DIF).

ShinyItemAnalysis covers broad range of methods and offers data examples, model equations, parameter estimates, interpretation of results, together with selected R code, and is thus suitable for teaching psychometric concepts with R. Besides, the application aspires to be a simple tool for analysis of educational tests and other composite measurements by allowing the users to upload and analyze their own data and to automatically generate analysis report in PDF or HTML.

The R package difNLR has been developed for detection of potentially unfair items in educational and psychological testing, analysis of so called differential item functioning, based on extensions of logistic regression model. For dichotomous data, six models have been implemented to offer wide range of proxies to Item Response Theory models. Parameters are obtained using non-linear least square estimation and DIF detection procedure is performed by either F or likelihood ratio test of submodel. For unscored data, analysis of differential distractor functioning (DDF) based on multinomial regression model is offered to provide closer look at individual item options (distractors).

We argue that psychometric analysis should be a routine part of test development in order to gather proofs of reliability and validity of the measurement. With example of admission test to medical school we demonstrate how presented R packages and Shiny application may provide simple and free tools to routinely analyze tests and to explain advanced psychometric models to students and to test developers. Attention will also be paid to technical details of automatically generated reports. 

Keywords: psychometrics, item response theory, detection of item bias, differential item functioning, shiny, R

PřednášejícíPatrícia Martinková, Adéla Drabinová, Jakub Houdek
Místo: Učebna č. 222, Ústav informatiky AV ČR, v.v.i., Pod Vodárenskou věží 2, 182 07 Praha 8
Datum: Čtvrtek 29. června 2017
Čas: 14:00 hod

Brain networks and scalp electroencephalogram

přidáno: 13. 4. 2017 2:58, autor: Zdeněk Valenta   [ aktualizováno 13. 4. 2017 3:02 ]

Milan Paluš
Oddělení nelineární dynamiky a složitých systémů, Ústav informatiky AV ČR v.v.i., Pod Vodárenskou věží 2, 182 07 Praha 8

AbstractUnderstanding how neurons and neuronal assemblies communicate is one of the greatest challenges of modern science. Adequate description and quantification of brain connectivity (i.e., communication between neuronal assemblies) is not only important for understanding the structure and function of brain networks, but also for diagnosis and treatment of neuropsychiatric diseases, since brain disorders – from schizophrenia to depression to post-traumatic stress disorder – are considered as disorders of connectivity. Functional brain networks are derived from multivariate time series of a quantity reflecting time evolution of brain activity. Modern neuroimaging methods became popular for the inference of the functional networks, however, the scalp electroencephalogram (EEG) is probably the most available and least expensive, non-invasive method to record the brain electrical activity.  We will discuss measures of synchronization and coherence which can be used to infer connectivity patterns from scalp EEG, with a special emphasis on measures designed to cope with the effects of conductivity and reference electrode. Another challenging topic is the detection of cross-frequency interactions, namely the phase-amplitude coupling. We will ask whether we can detect cross-frequency interactions from scalp EEG and whether we can identify them as causal, e.g. in the sense “the phase of slow oscillations determines the amplitude of fast oscillations.” 
 
KeywordsEEG, time series analysis, brain networks, functional connectivity, synchronization, coherence, cross-frequency phase-amplitude coupling

PřednášejícíRNDr. Milan Paluš, DrSc.
Místo: Učebna č. 222, Ústav informatiky AV ČR, v.v.i., Pod Vodárenskou věží 2, 182 07 Praha 8
Datum: Čtvrtek 18. května 2017
Čas: 13:00 hod

Some applications of numerical linear algebra in statistics and some applications of statistics in pharmacy

přidáno: 2. 3. 2017 3:57, autor: Zdeněk Valenta   [ aktualizováno 8. 3. 2017 13:28 ]

Duintjer Jurjen Tebbens
Oddělení výpočetních metod, Ústav informatiky AV ČR v.v.i., Pod Vodárenskou věží 2, 182 07 Praha 8

AbstractThe talk will present two kinds of personal experience: In the first place, personal experience with attempts to apply knowledge and skills from numerical linear algebra (matrix computations), my primary field of research, to some selected statistical problems. The experiences were gathered mainly while co-authoring papers on classification with discriminant analysis and on robust multivariate scatter estimation. In the second place, I will report on personal experience and future work in solving statistical tasks that arise in pharmaceutical research. This second part does not present any original research, its intention is rather to inform the audience about an application area that is perhaps not very well known to statisticians. 
 
KeywordsSparse matrices, computational and memory costs, low rank matrices, Fisher’s linear discriminant analysis, the minimum covariance determinant estimator, design of experiments, meta-analysis.

Přednášející: Dipl. Math. Duintjer Jurjen Tebbens, PhD.
Místo: Učebna č. 222, Ústav informatiky AV ČR, v.v.i., Pod Vodárenskou věží 2, 182 07 Praha 8
Datum: Čtvrtek 4. května 2017
Čas: 13:00 hod

Měření a klasifikace fotopletysmografických signálů a jejich potenciál pro diagnostiku kardiovaskulárních onemocnění

přidáno: 24. 2. 2017 1:46, autor: Zdeněk Valenta   [ aktualizováno 2. 3. 2017 5:37 ]

Karel Kupka
TriloByte Statistical Software, Ltd., TBSA: TriloByte Statistical Academy, Staré Hradiště 300, CZ 53352 Pardubice

Hemodynamics data carries information about both central and peripheral arterial system. Photoplethysmography (PPG) is widely used as a simple diagnostic tool. In this contribution we investigate some methods of statistical modelling including smoothing, orthogonal harmonic regression, parametrization of the PPG pulses and clustering and classification of different curve shapes in the multivariate parametric space to aid diagnosis. We also investigate alternative method called baroplethysmography (BPG) which records blood pressure signal. Both methods are non-invasive and are applied on patient’s finger. We conclude that both measurement methods can be applied simultaneously and the measured signal may provide information transormed into multiple (typically 10-20) independent parameters which are used both to long-term stability monitoring of a particular patient and to classify different patients with respect to their arterial system and possible related diseases.
 
Keywords: Plethysmography, Haemodynamics, Arterial system diagnosis, Signal processing, Harmonic regression, Clustering, Classification, Support Vector Machine

Přednášející: Ing. Karel Kupka, PhD.&PhD.
Místo: Učebna č. 222, Ústav informatiky AV ČR, v.v.i., Pod Vodárenskou věží 2, 182 07 Praha 8
Datum: Čtvrtek 30. března 2017
Čas: 13:00 hod

Před přednáškou proběhne od 12:30 hod Valná hromada společnosti, kde budou předneseny zprávy o činnosti a hospodaření za uplynulý rok 2016, včetně plánu činnosti pro rok 2017.

Bayesovské metody ve farmaceutické výrobě

přidáno: 22. 3. 2016 3:13, autor: Zdeněk Valenta   [ aktualizováno 22. 3. 2016 9:44 ]

Martin Otava
Nonclinical Statistics and Computing, Janssen Research & Development, Janssen Pharmaceutica NV, Turnhoutseweg 30, B-2340 Beerse, Belgium

Jedním z typických statistických problémů ve farmaceutickém výrobním procesu je vyhodnocení budoucího experimentu, který formálně ověří kvalitu daného procesu vzhledem ke standardizovaným kritériím. Pravděpodobnost „úspěchu“ takového experimentu se pak vypočítává na základě dat získaných během předchozích stupňů vývoje. Vzhledem k tomu, že mnoho kritérií se vztahuje na individuální pozorování, častým řešením je odhad prediktivních intervalů či tolerančních intervalů (které berou v potaz variabilitu bodových odhadů střední hodnoty a rozptylu). Alternativním postupem je Bayesovská analýza problému a přímý odhad rozdělení individuálních hodnot. Výhodou je snadné zahrnutí náhodných efektů (často z několika různých zdrojů variability) a především snadná interpretace výstupu analýzy, kterým je pravděpodobnost úspěchu daného budoucího experimentu. 

Přednášející: Mgr. Martin Otava, Ph.D. (Janssen Pharmaceutica NV, Belgie)
Místo: Praktikum KPMS (1. patro vedle schodů), MFF UK, Sokolovská 83
Datum: Středa 27. dubna 2016
Čas: 15:40 hod

Modeling risk of orthopedic implant failure using kernel estimation

přidáno: 11. 2. 2016 7:08, autor: Zdeněk Valenta   [ aktualizováno 2. 5. 2016 7:15 ]

Selingerová Iveta, Katina Stanislav, Zelinka Jiří, Horová Ivana
Ústav matematiky a statistiky
Přírodovědecká fakulta 
Masarykova universita Brno
Kotlářská 267/2
611 37 Brno

The hazard function is an important tool in survival analysis and reflects the instantaneous probability of failure occurrence within the next time instant. The hazard function can depend on any covariates as age, gender, etc. We use method of kernel smoothing for modeling of the unconditional and conditional hazard function. We include estimates using Cox proportional hazard model smoothed by kernel methods as well. Attention is also paid to comparing both approaches. These methods are applied to the real data from Slovak Arthroplasty Register about implants of an artificial hip joint replacement implemented in all 40 orthopedic and traumatology departments in the Slovak Republic (coverage of 99.9%) with a maximum duration of follow-up of twelve years from Jan 1 2003 to Dec 31 2014. The set of 46 859 operations with 1005 implant failures is stratified based on types of fixation, diagnosis, and gender. The hazard function conditioned on age in years is calculated for pre-specified data-subsets and visualized as color-coded surfaces. These results will lead to an improvement of the quality of care for patients after artificial joint replacements.

KeywordsConditional hazard function, Kernel estimation, Cox regression model, Orthopedic data, the Slovak Arthroplasty Register


PřednášejícíDoc. PaedDr. RNDr. Stanislav Katina, Ph.D.Ústav matematiky a statistiky, Přírodovědecká fakulta Masarykovy university v Brně, Kotlářská 267/2, 611 37 Brno

Datum: Čtvrtek 5. května 2016

Čas: 13:30 hod

1-10 of 35