Stefan Kemeny

Metabolic and electrophysiological validation of functional MRI

OBJECTIVES Although functional MRI is widely used for preoperative planning and intraoperative neuronavigation, its accuracy to depict the site of neuronal activity is not exactly known. Experience with methods that may validate fMRI data and the results obtained when coregistering fMRI with different preoperative and intraoperative mapping modalities including metabolically based 18F-fluorodeoxyglucose PET, electrophysiologcally based transcranial magnetic stimulation (TMS), and direct electrical cortical stimulation (DECS) are described. METHODS Fifty patients were included. PET was performed in 30, TMS in 10, and DECS in 41 patients. After coregistration using a frameless stereotactic system, results were grouped into overlapping (<1 cm distance), neighbouring (<2 cm), or contradictory (>2 cm). RESULTS Comparing fMRI with PET, 18 overlapping, seven neighbouring, and one contradictory result were obtained. In four patients no comparison was possible (because of motion artefacts, low signal to noise ratio, and unusual high tumour metabolism in PET). The comparison of TMS and fMRI showed seven overlapping and three neighbouring results. In three patients no DECS results could be obtained. Of the remaining 38 patients, fMRI hand motor tasks were compared with DECS results of the upper limb muscles in 36 patients, and fMRI foot motor tasks were compared with DECS results of the lower limb on 13 occasions. Of those 49 studies, overlapping results were obtained in 31 patients, and neighbouring in 14. On four occasions fMRI did not show functional information (because of motion artefacts and low signal to noise). CONCLUSIONS All validation techniques have intrinsic limitations that restrict their spatial resolution. However, of 50 investigated patients, there was only one in whom results contradictory to fMRI were obtained. Although it is not thought that fMRI can replace the intraoperatively updated functional information (DECS), it is concluded that fMRI is an important adjunct in the preoperative assessment of patients with tumours in the vicinity of the central region.

Asymmetries of visual attention after circumscribed subcortical vascular lesions.

OBJECTIVE To investigate the role of the basal ganglia and the thalamus for basic processes of visuospatial attention METHODS Fifteen patients with acute circumscribed vascular lesions (10 with haemorrhage and five with infarction) were included in the study. The lesions were confined exclusively to subcortical structures, such as the basal ganglia, internal capsule, and thalamus, which was confirmed by initial CT on the day of referral and MRI taken 14–28 days after clinical onset. These patients were examined with two computerised attentional tasks (one detection and one search task) measuring spatial visual attention. RESULTS There was a clear attentional asymmetry in patients with right hemispheric lesions (RHLs) in the visual search task. Seven out of eight patients with RHLs tended to be slower and/or missed significantly more target stimuli in the left sided part of a stimulus array consisting of 25 small squares than in right sided parts, although none of these patients showed signs of visual hemineglect in the visual detection task presenting visual information simultaneously to the right and left visual hemispace. All but one of these patients showed lesions in the posterior limb of the internal capsule and the putamen. On the other hand, patients with left hemispheric lesions were not impaired in the search task with only one patient showing more contralesional omissions of target stimuli than could be expected from the behaviour of normal controls. CONCLUSIONS The results are in line with previous results showing a dominant role of right hemispheric neuronal structures for spatial attention. Furthermore, the data suggest that even with right hemispheric subcortical lesions without cortical involvement deficits in spatial orienting of attention to the left hemispace can be seen. These asymmetries of visual attention in the absence of neglect symptoms are supposed to be caused (1) by a disruption of the motor corticostriato-pallidothalamo-cortical neuronal circuit or (2) by a (partial) disconnection of relevant parts within the posterior attention network—namely, parietal and thalamic structures.

The representation of the plegic hand in the motor cortex: a combined fMRI and TMS study.

TMS mapping and fMRI were used to investigate changes in the motor cortex representation of the hand in a patient with complete loss of right hand function following traumatic avulsion of the cervical roots C7 and C8. Both TMS and fMRI demonstrated an expansion of the motor representation of the forearm into the hand area contralateral to the injured side. fMRI of the hand area, however, revealed that this area could still be activated when the patient was instructed to imagine finger tapping with his plegic hand. These results indicate that the plegic hand is still represented in the motor cortex, despite the fact that the same cortical area is also now active during movements involving forearm muscles.

Recovery of semantic word processing in transcortical sensory aphasia

In a previous functional magnetic resonance imaging (fMRI) study with normal subjects, we demonstrated regions related to conceptual-semantic word processing around the first frontal sulcus (BA 9) and the posterior parietal lobe (BA 7/40) in agreement with several previous reports. We had the possibility, using the same fMRI paradigm, to study two consecutive cases with left middle cerebral artery (MCA) infarction (RC and HP) and lesions affecting either solely the pre-frontal (HP) or both the pre-frontal and posterior parietal part of the network activated in normal subjects (RC). Both patients showed transcortical sensory aphasia (TSA) on acute assessment. This contradicts classical disconnection accounts of the syndrome stating intact conceptual representations in TSA. Their recovery of language comprehension was associated with activation of a left hemispheric network. Mainly activations of left perilesional pre-frontal regions (RC), left Wernicke’s area (RC and HP) or the left posterior middle and inferior temporal cortex (HP) were demonstrated in the TSA patients. The latter findings suggest that in our cases of TSA functional take-over has occurred in regions with related functions (‘redundancy recovery’) rather than in previously unrelated areas (‘vicarious functioning’). Our data support distributed models of conceptual-semantic word processing and multiple left hemispheric representations of closely related functions.

Unsupervised time course analysis of functional magnetic resonance imaging (fMRI) using self-organizing maps (SOMs)

Functional Magnetic Resonance Imaging (fMRI) data of the brain includes activated parenchymal voxels, corresponding to the paradigm performed, non-activated parenchymal voxels and background voxels. Statistical tests, e.g. using the general linear model approach of SPM or the Kolmogorov-Smirnov (KS) non-parametric statistic, are common supervised techniques to look for activation in functional brain MRI. Selection of voxel type by comparing the voxel time course with a model of the expected hemodynamic response function (HRF) from the task paradigm has proven to be difficult due to individual and spatial variance of the measured HRF. For the functional differentiation of brain voxels we introduce a method separating brain voxels based on their features in the time domain using a self-organizing map (SOM) neural network technique without modeling the HRF. Since activation measured by fMRI is related to magnetic susceptibility changes in venous blood which represents only 2 - 5% of brain matter, preprocessing is required to remove the majority of non- activated voxels which dominate learning instead of real activation patterns. Using the auto-correlation function one can select voxels which are candidates of being activated. Features of the time course of the selected voxels can be learned with the SOM. In the first step the SOM is trained by the voxels time course, fitting its neurons to the input. After learning, the neurons have adapted to the intrinsic features space of the voxel time courses. Using the trained SOM, voxel time courses are presented again, now labeled by the neuron having the smallest Euclidean distance to the presented voxel time course. The result of the labeling and the learned feature time course vectors are compared visually with the p-value map of the KS statistic. With the SOM map one can visually separate the voxels based on their features in the time domain into different functional task related classes.

Analysis of short single rest/activation epoch fMRI by self-organizing map neural network

Functional magnet resonance imaging (fMRI) has become a standard non invasive brain imaging technique delivering high spatial resolution. Brain activation is determined by magnetic susceptibility of the blood oxygen level (BOLD effect) during an activation task, e.g. motor, auditory and visual tasks. Usually box-car paradigms have 2 - 4 rest/activation epochs with at least an overall of 50 volumes per scan in the time domain. Statistical test based analysis methods need a large amount of repetitively acquired brain volumes to gain statistical power, like Students t-test. The introduced technique based on a self-organizing neural network (SOM) makes use of the intrinsic features of the condition change between rest and activation epoch and demonstrated to differentiate between the conditions with less time points having only one rest and one activation epoch. The method reduces scan and analysis time and the probability of possible motion artifacts from the relaxation of the patients head. Functional magnet resonance imaging (fMRI) of patients for pre-surgical evaluation and volunteers were acquired with motor (hand clenching and finger tapping), sensory (ice application), auditory (phonological and semantic word recognition task) and visual paradigms (mental rotation). For imaging we used different BOLD contrast sensitive Gradient Echo Planar Imaging (GE-EPI) single-shot pulse sequences (TR 2000 and 4000, 64 X 64 and 128 X 128, 15 - 40 slices) on a Philips Gyroscan NT 1.5 Tesla MR imager. All paradigms were RARARA (R equals rest, A equals activation) with an epoch width of 11 time points each. We used the self-organizing neural network implementation described by T. Kohonen with a 4 X 2 2D neuron map. The presented time course vectors were clustered by similar features in the 2D neuron map. Three neural networks were trained and used for labeling with the time course vectors of one, two and all three on/off epochs. The results were also compared by using a Kolmogorov-Smirnov statistical test of all 66 time points. To remove non- periodical time courses from training an auto-correlation function and bandwidth limiting Fourier filtering in combination with Gauss temporal smoothing was used. None of the trained maps, with one, two and three epochs, were significantly different which indicates that the feature space of only one on/off epoch is sufficient to differentiate between the rest and task condition. We found, that without pre-processing of the data no meaningful results can be achieved because of the huge amount of the non-activated and background voxels represents the majority of the features and is therefore learned by the SOM. Thus it is crucial to remove unnecessary capacity load of the neural network by selection of the training input, using auto-correlation function and/or Fourier spectrum analysis. However by reducing the time points to one rest and one activation epoch either strong auto- correlation or a precise periodical frequency is vanishing. Self-organizing maps can be used to separate rest and activation epochs of with only a 1/3 of the usually acquired time points. Because of the nature of the SOM technique, the pattern or feature separation, only the presence of a state change between the conditions is necessary for differentiation. Also the variance of the individual hemodynamic response function (HRF) and the variance of the spatial different regional cerebral blood flow (rCBF) is learned from the subject and not compared with a fixed model done by statistical evaluation. We found that reducing the information to only a few time points around the BOLD effect was not successful due to delays of rCBF and the insufficient extension of the BOLD feature in the time space. Especially for patient routine observation and pre-surgical planing a reduced scan time is of interest.

Perception of Signs and Written Words: An Fmri Study

PERCEPTION OF SIGNS AND WRITTEN WORDS: AN fMRI STUDY Juliane Klann1, Frank Kastrau2, Stefan Kémeny2 and Walter Huber1 (1Neurolinguistics and 2Neurology at the Department of Neurology, University Hospital Aachen, Germany)

Wissensbasierte Optimierung von Selbstorganisierenden Merkmalskarten (SOM) zur Analyse von funktionellen Magnetresonanztomographien (fMRT)

Die funktionelle Magnetresonanztomographie (fMRT) des Gehirns ermoglicht die Lokalisation funktioneller Ablaufe im Gehirn. Erhother Blutflus als Folge von erhohter neuronaler Aktivitat in einem aktivierten Hirnareal ist mit Hilfe sehr schneller MR-Bildgebung mesbar. Durch geeignete Aktivierungsparadigmen uber einen Zeitraum ist es moglich, aktivierte von nicht aktivierten Regionen zu trennen. Zur Trennung werden die gemessenen Zeitreihen (MR-Signal in der Zeit) fur jedes Voxel auf dessen Korrespondenz mit dem Aktivierungsparadigma uberpruft. Neben den bisher verwendeten statistischen Verfahren ist es auch moglich eine Separierung der Zeitreihen mittels einer selbstorganisierenden Merkmalskarte (SOM, Self- Organizing Map) zu erreichen, ohne dabei ein Modell der hamodynamischen Antwort des Gehirns zu verwenden. Die nur sehr geringe Anzahl an aktivierten Voxel setzt eine Vorauswahl geeigneter Kandidaten fur das Training des SOM Neuronalen Netzwerkes voraus. Die hier beschriebene Methode wahlt die geeignetsten Lernkandidaten anhand des Vorwissens uber die Periodizitat des Aufgabenparadigmas unter Verwendung des Fourierspektrums bzw. des Frequenzperiodogramms jeder Zeitreihe aus, wodurch eine erhebliche Verbesserung der Klassifikationsleistung der SOM erreicht wird.

Unüberwachte Zeitreihenanalyse von fMRT-Daten

Funktionelle Untersuchungen des Gehirns mittels schneller Echo-Planar Magnet-Resonanz-Tomographie (fMRT) liefern ortliche und zeitliche Information uber komplexe Hirnaktivitaten. Die vorgestellte Methode versucht, eine Zerlegung der gemessenen Zeitreihen anhand ihrer charakteristischen Merkmale in disjunkte Klassen, d.h. echte Aktivierung, ruhendes Gewebe, Hintergrund oder Bewegungsartefakt zu gewinnen. Die Zerlegung erfolgt mittels eines neuronalen Netzwerks, der selbst-organisierenden Merkmalskarte (SOM), die unuberwacht, d.h. ohne zusatzliche Informationen uber das Aufgabenparadigma oder einer Modellierung der hamodynamischen Antwort, die hochdimensionalen Zeitreihen in eine 2D Neuronenkarte uberfuhrt. Ziel ist eine den Klassen entsprechende geordnete Bildung von Clustern auf der Neuronenkarte, die als Eingabe fur einen anschliesenden Klassifikationsprozes geeignet ist. Es zeigt sich, das die SOM in der Lage ist, Cluster von ahnlichen anatomischen oder funktionellen Voxeln zu bilden und diese Ahnlichkeit auch raumlich in der Neuronenkarte abzubilden.