György Kozmann

Evaluation of the brain network organization from EEG signals: a preliminary evidence in stroke patient.

Synchronous brain activity in motor cortex in perception or in complex cognitive processing has been the subject of several studies. The advanced analysis of cerebral electro‐physiological activity during the course of planning (PRE) or execution of movement (EXE) in a high temporal resolution could reveal interesting information about the brain functional organization in patients following stroke damage. High‐power (128 channels) electroencephalography registration was carried out on 8 healthy subjects and on a patient with stroke with capsular lacuna in the right hemisphere. For activation of motor cortex, the finger tapping paradigm was used. In this preliminary study, we tested a theoretical graph approach to characterize the task‐related spectral coherence. All of the obtained brain functional networks were analyzed by the connectivity degree, the degree distribution, and efficiency parameters in the Theta, Alpha, Beta, and Gamma bands during the PRE and EXE intervals. All the brain networks were found to hold a regular and ordered topology. However, significant differences (P < 0.01) emerged between the patient with stroke and the control subjects, independently of the neural processes related to the PRE or EXE periods. In the Beta (13–29 Hz) and Gamma (30–40 Hz) bands, the significant (P < 0.01) decrease in global‐ and local‐efficiency in the patients networks, reflected a lower capacity to integrate communication between distant brain regions and a lower tendency to be modular. This weak organization is principally due to the significant (P < 0.01 Bonferroni corrected) increase in disconnected nodes together with the significant increase in the links in some other crucial vertices. Anat Rec, 292:2023–2031, 2009.

The use of the SPSA method in ECG analysis

The classification, monitoring, and compression of electrocardiogram (ECG) signals recorded of a single patient over a relatively long period of time is considered. The particular application we have in mind is high-resolution ECG analysis, such as late potential analysis, morphology changes in QRS during arrythmias, T-wave alternants, or the study of drug effects on ventricular activation. We propose to apply a modification of a classical method of cluster analysis or vector quantization. The novelty of our approach is that we use a new distortion measure to quantify the distance of two ECG cycles, and the class-distortion measure is defined using a min-max criterion. The new class-distortion-measure is much more sensitive to outliers than the usual distortion measures using average-distance. The price of this practical advantage is that computational complexity is significantly increased. The resulting nonsmooth optimization problem is solved by an adapted version of the simultaneous perturbation stochastic approximation (SPSA) method of J. Spall (IEEE Trans. Automat. Contr., vol. 37, p. 332-41, Mar. 1992). The main idea is to generate a smooth approximation by a randomization procedure. The viability of the method is demonstrated on both simulated and real data. An experimental comparison with the widely used correlation method is given on real data.

SPSA for non-smooth optimization with application in ECG analysis

It is shown that nonsmooth optimization problems can be solved by a suitable extension of the simultaneous perturbation stochastic approximation (SPSA) method. The new optimization method has been tested in a min-max classification problem using both simulated and real data. The latter are ECG signals which were collected for the detection of so-called late potentials.

Body surface potential field representation fidelity: Analysis of map estimation procedures

The first part of this study analyzed the spatial-temporal error distribution of the Lux-type limited lead system. Quantitative new evidence is reported that the 32-lead anterior subset estimates the further 160 leads with an average amplitude error less than 38.5 microV. The spatial error distribution revealed 8 sites where the error is the highest, primarily on the anterior side, independent of the clinical classification. The second part of the study examined inter-lead-system conversion strategies for interpolating the Lux-192 lead maps from the Montreal-63 data. The methodology based on the Laplacian interpolation yielded an average amplitude error of 143.7 microV and an average correlation of 0.87 for pattern fidelity. In this specific case a modified linear interpolation surpassed the Laplacian method. A presented example illustrates that even in cases when the fidelity of the signal information is heavily compromised the diagnostic information may remain less influenced.

Application of Artificial Intelligence for Weekly Dietary Menu Planning

Dietary menu planning is an important part of personalized lifestyle counseling. The chapter describes the results of an automated menu generator (MenuGene) of the web-based lifestyle counseling system Cordelia that provides personalized advice to prevent cardiovascular diseases. The menu generator uses Genetic Algorithms to prepare weekly menus for web users. The objectives are derived from personal medical data collected via forms, combined with general nutritional guidelines. The weekly menu is modeled as a multi-level structure. Results show that the Genetic Algorithm based method succeeds in planning dietary menus that satisfy strict numerical constraints on every nutritional level (meal, daily basis, weekly basis). The rule-based assessment proved capable of manipulating the mean occurrence of the nutritional components thus providing a method for adjusting the variety and harmony of the menu plans. By splitting the problem into well determined subproblems, weekly menu plans that satisfy nutritional constraints and have well assorted components can be generated with the same method that is used for daily and meal plan generation.

Methods to highlight consistency in repeated EEG recordings

Abstract In the present work, we aimed to find subject related features in EEG recordings which are consistent through multiple recordings and apply them in biometry. Essentially to use the brains electroencephalographic activity as a possible way to identify individuals. Seventeen healthy subjects participated in the study and their brain activity were recorded through a 56 EEG channel, high-density EEG cap during one minute of resting state with eyes open and/or eyes closed. The subjects were participating in a second recording session as well, thus creating a dataset of ten closed and ten open eyed recordings each with follow-up measurements. Analyzing results of various testing scenarios involving power spectrum density (PSD), coherence (COH), and the imaginary part of coherence (iCOH) on segments of ten seconds, we concluded the best parameter setup as well as a minimal set of electrodes and the best possible feature vector assembly based on these computations. By using a naive Bayes classifier and K-fold cross-validations, we observed the highest correct recognition rates (CRR 98.33%) during eyes closed resting state at the parieto-occipital-temporal electrodes, suggesting these as the most stable characteristics therefore the most suitable, among those investigated here, for identifying individuals.

Personalized Nutrition Counseling Expert System

This paper gives an overview of the MenuGene nutrition counseling expert system and its main components, focusing on home health monitoring / dietary logging. Pub- lished solutions so far can satisfy nutrient constraints, but does not take harmony rules into account. The novelty of the algo- rithm used by MenuGene is that dietetic knowledge including harmony rules can take part of the menu synthesis process. Our results show that a considerable part of dietary counseling can be extended by computer aided approach, if we provide easy to use interfaces for patients and dietetic experts. Accord- ing to our tests with volunteers an average logging accuracy of ca. 15% can be achieved with our Dietlog web interface.

A formal domain model for dietary and physical activity counseling

Diet and physical activity planning is a complex process that usually comprises repetitive expert-patient discussions and multihour construction phases. Recent advances in artificial intelligence and improvements in CPU speeds make it now possible to enhance or even substitute the work of the dietary expert. Although research in this field began as early as the 1940s, no comprehensive domain model has been developed to date. Previous works reduced the problem to then solvable mathematical models, thus lessening the quality of the solution. Here, we present a novel domain model which can handle the multi-objective nature of the problem as well as the proper use of expert knowledge on dietary harmony. The model provides a base for the computerized planning of human-competetive solutions. An implementation of this model is employed in the nutrition and lifestyle counseling expert system Menugene.

An evolutionary divide and conquer method for long-term dietary menu planning

We present a novel Hierarchical Evolutionary Divide and Conquer method for automated, long-term planning of dietary menus. Dietary plans have to satisfy multiple numerical constraints (Reference Daily Intakes and balance on a daily and weekly basis) as well as criteria on the harmony (variety, contrast, color, appeal) of the components. Our multi-level approach solves problems via the decomposition of the search space and uses good solutions for sub-problems on higher levels of the hierarchy. Multi-Objective Genetic Algorithms are used on each level to create nutritionally adequate menus with a linear fitness combination extended with rule-based assessment. We also apply case-based initialization for starting the Genetic Algorithms from a better position of the search space. Results show that this combined strategy can cope with strict numerical constraints in a properly chosen algorithmic setup.

A GPU-based simultaneous real-time EEG processing and visualization system for brain imaging applications

A data-driven prototype software is presented for EEG processing and visualization. The system relies on the GPU architecture for providing simultaneous processing and visualization of the EEG data. Two example brain imaging algorithms, the surface Laplacian and the spherical forward solution are used for illustrating the effective use of the massively parallel GPU hardware in speeding up computations. The paper describes the architecture of our system, the key design decisions, and the performance optimization of the parallel implementation. Using the CUDA-OpenGL interoperability, the computing subsystem can directly modify potential data in the OpenGL vertex memory, avoiding unnecessary GPU-Host data transfers. The system and our parallel implementations demonstrate that real-time processing and visualization is possible for a range of algorithms during EEG processing. We are confident that these results can pave the way for supercomputing-class implementations and open up new opportunities in the clinical practice and neuroscience research.