Axel Baune

Dynamical Cluster Analysis of Cortical fMRI Activation

Localized changes in cortical blood oxygenation during voluntary movements were examined with functional magnetic resonance imaging (fMRI) and evaluated with a new dynamical cluster analysis (DCA) method. fMRI was performed during finger movements with eight subjects on a 1.5-T scanner using single-slice echo planar imaging with a 107-ms repetition time. Clustering based on similarity of the detailed signal time courses requires besides the used distance measure no assumptions about spatial location and extension of activation sites or the shape of the expected activation time course. We discuss the basic requirements on a clustering algorithm for fMRI data. It is shown that with respect to easy adjustment of the quantization error and reproducibility of the results DCA outperforms the standard k-means algorithm. In contrast to currently used clustering methods for fMRI, like k-means or fuzzy k-means, DCA extracts the appropriate number and initial shapes of representative signal time courses from data properties during run time. With DCA we simultaneously calculate a two-dimensional projection of cluster centers (MDS) and data points for online visualization of the results. We describe the new DCA method and show for the well-studied motor task that it detects cortical activation loci and provides additional information by discriminating different shapes and phases of hemodynamic responses. Robustness of activity detection is demonstrated with respect to repeated DCA runs and effects of different data preprocessing are shown. As an example of how DCA enables further analysis we examined activation onset times. In areas SMA, M1, and S1 simultaneous and sequential activation (in the given order) was found.

Object Classification Using Simple, Colour Based Visual Attention and a Hierarchical Neural Network for Neuro-symbolic Integration

An object classification system built of a simple colour based visual attention method, and a prototype based hierarchical classifier is established as a link between subsymbolic and symbolic data processing. During learning the classifier generates a hierarchy of prototypes. These prototypes constitute a taxonomy of objects. By assigning confidence values to the prototypes a classification request may also return symbols with confidence values. For performance evaluation the classifier was applied to the task of visual object categorization of three data sets, two real-world and one artificial. Orientation histograms on subimages were utilized as features.With the currently very simple feature extraction method, classification accuracies in the range of 69% to 90% were attained.

Object classification with simple visual attention and a hierarchical neural network for subsymbolic-symbolic coupling

An object classification system using a simple color based visual attention method, and a prototype based hierarchical classifier is established as a link between sub-symbolic and symbolic data processing. During learning the classifier generates a hierarchy of prototypes. These prototypes constitute a taxonomy of objects. By assigning confidence values to the prototypes a classification request may also return symbols with confidence values. For performance evaluation the classifier was applied to the task of visual object categorization of three data sets, two real-world and one artificial. Orientation histograms on sub-images were utilized as features. With the currently very simple feature extraction method, classification accuracies of about 75% to 90% were attained.

Improving binding potential analysis in [11C]raclopride PET studies using cluster analysis

To calculate binding potentials (BP) in [11C]raclopride brain PET studies a reference tissue model is widely used. The aim of the present study was to improve the determination of time activity curves (TAC) of reference tissue regions using cluster analysis. In four patients with Huntington disease TACs of a cerebellar reference region were calculated either from manually placed circular ROIs within the cerebellum or by cluster analysis. BP estimates derived from cluster analysis are independent from inter- and intraobserver variations and show an improved reproducibility combined with a low variability compared to manually placed cerebellar ROIs. This is of high value in longitudinal studies.

Dynamical Cluster Analysis for the Detection of Microglia Activation

Dynamical cluster analysis (DCA) was used to extract sets of representative time courses to detect brain lesions using positron emmision tomography (PET) data. DCA is an adaptive hard-clustering algorithm where the number of clusters k is not initially fixed but is dynamically changed by generation and fusion of clusters during runtime. We analyzed PET data sets of 9 patients applying DCA repeatedly. We compared the results that vary in the number of clusters even on the same data set. As validation measure we used the mean square quantization error (MSQE). We found that the MSQE was strictly correlated with k only on 4 of the 9 data sets. We propose DCA for extracting the reference time course required in reference tissue modeling [7]. In the case of one patient, we checked the ability of DCA to characterize directly the three most interesting regions, reference tissue, the veins and the lesion and how this ability relates to high validation scores. The characterisation of all three regions was not reproducible in all of the runs, however, runs rated high in validity by the MSQE were able to reproduce all the three regions.

A Hierarchical Neural Object Classifier for Subsymbolic-Symbolic Coupling

A prototype based hierarchical classifier is established as a link between subsymbolic and symbolic data processing. During learning the classifier generates a hierarchy of prototypes. These prototypes constitute a taxonomy of objects. By assigning confidence values to the prototypes a classification request may also return symbols with confidence values.

Interpretation of Event-Related fMRI Using Cluster Analysis

Event-related fMRI can access more complex experimental paradigms than traditional block designs. A new problem, however, becomes the data analysis, i.e., the generation of appropriate statistical models. For parametric paradigms where a set of parameters describing task and stimulus can be varied, the number of different conditions becomes large. Even if the combinatorics of contrasts between conditions is reduced by assumptions, the number of contrasts that potentially may contribute to the functional interpretation becomes huge.

Dynamische Clusteranalyse von [ 11 C]-PK11195 PET Daten des Gehirns

Diese Studie untersucht die Moglichkeit, Bereiche unterschiedlicher Biokinetik durch dynamische Clusteranalyse auf [11C]-PK11195 PET Daten automatisch zu diskriminieren. Die dynamische Clusterung wird verglichen mit einer manuellen Segmentierung aufgrund visueller Inspektion. Auserdem wird das vorgeschlagene automatische Verfahren gegenuber dem klassischen k-means Verfahren abgegrenzt.

Schritthaltende Objektklassifikation für einen autonomen mobilen Roboter

Der hier vorgestellte Ansatz zur schritthaltenden Objektklassifikation in einem autonomen mobilen Robotersystem wurde im Rahmen des Sonderforschungsbereichs Integration symbolischer und subsymbolischer Informationsverarbeitung in adaptiven sensormotorischen Systemen (SFB 527) entwickelt. Ziel dieses SFBs ist die Erforschung neuronaler und wissensbasierter Methoden zur symbolischen und subsymbolischen Informationsverarbeitung, sowie die Kombination und Integration dieser Verfahren auf einem autonomen mobilen Robotersystem [1]. Dieser Roboter soll sich in einer Buroumgebung zurecht finden, auf interaktiv gestellte Aufgaben reagieren konnen, beispielsweise Objekte suchen und einsammeln, Personen suchen und fuhren und mit Personen kommunizieren, sowie auf unerwartete Ereignisse in der Umwelt reagieren. Fur fast alle Aufgaben ist eine robuste und moglichst schnelle Objekterkennung unumganglich. Im folgenden wird der von uns implementierte Ansatz zur visuellen Objekterkennung beschrieben.