A BCI using VEP for continuous control of a mobile robot.

A brain-computer interface (BCI) translates brain activity into commands to control devices or software. Common approaches are based on visual evoked potentials (VEP), extracted from the electroencephalogram (EEG) during visual stimulation. High information transfer rates (ITR) can be achieved using (i) steady-state VEP (SSVEP) or (ii) code-modulated VEP (c-VEP). This study investigates how applicable such systems are for continuous control of robotic devices and which method performs best. Eleven healthy subjects steered a robot along a track using four BCI controls on a computer screen in combination with feedback video of the movement. The average time to complete the tasks was (i) 573.43 s and (ii) 222.57 s. In a second non-continuous trial-based validation run the maximum achievable online classification accuracy over all subjects was (i) 91.36 % and (ii) 98.18 %. This results show that the c-VEP fits the needs of a continuous system better than the SSVEP implementation.

Performance of a tactile P300 speller for healthy people and severely disabled patients.

P300 based Brain-Computer Interfaces (BCIs) for communication are well known since many years. Most of them use visual stimuli to elicit evoked potentials because it is easy to integrate a high number of different classes into the paradigm. Nevertheless, a BCI that depends on visual stimuli is sometimes not feasible due to the presence of visual impairment in patients with severe brain injuries. In this case, it could be possible to use auditory or somatosensory stimulation. In this publication a vibrotactile P300 based BCI is introduced. Two different approaches were tested: a first approach using two stimulators and a second one that utilizes three stimulators for emitting the stimuli. The two paradigms were tested on 16 users: A group of ten healthy users and a second group comprising of 6 patients suffering Locked-In Syndrome. The control accuracy was calculated for both groups and both approaches, proving the feasibility of the device, not only for healthy people but also in severely disabled patients. In a second step we evaluated the influence of the number of stimuli on the accuracy. It was shown that in many cases the maximum accuracy was already reached with a small number of stimuli, this could be used in future tests to speed up the Information transfer rate.

Image processing pipeline for synchrotron-radiation-based tomographic microscopy.

With synchrotron-radiation-based tomographic microscopy, three-dimensional structures down to the micrometer level can be visualized. Tomographic data sets typically consist of 1000 to 1500 projections of 1024 x 1024 to 2048 x 2048 pixels and are acquired in 5-15 min. A processing pipeline has been developed to handle this large amount of data efficiently and to reconstruct the tomographic volume within a few minutes after the end of a scan. Just a few seconds after the raw data have been acquired, a selection of reconstructed slices is accessible through a web interface for preview and to fine tune the reconstruction parameters. The same interface allows initiation and control of the reconstruction process on the computer cluster. By integrating all programs and tools, required for tomographic reconstruction into the pipeline, the necessary user interaction is reduced to a minimum. The modularity of the pipeline allows functionality for new scan protocols to be added, such as an extended field of view, or new physical signals such as phase-contrast or dark-field imaging etc.

AAM-based segmentation for imaging cardiac electrophysiology.

OBJECTIVES: Activation time (AT) imaging from electrocardiographic (ECG) mapping data has been developing for several years. By coupling 4-dimensional volume data (3D + time) the electrical sequence can be computed non-invasively. In this paper an approach for extracting the ventricular and atrial blood masses for structurally normal hearts by using cine-gated short-axis data obtained via magnetic resonance imaging (MRI) is introduced. METHODS: The blood masses are extracted by employing Active Appearance Models (AAMs). The ventricular blood masses are segmented, applying the AAMs after providing apex cordis and base of the heart in the volume data, whereas the more complex geometry of the atria requires a more specific attempt. On account of this the atrium was divided into three divisions of appearance, where the images of the volume data in the related divisions have a maximum affinity. The first division reaches from the base of the heart to initial visibility of the upper and left lower pulmonary vein. The second division up from there to the last occurrence and the third division from there to the end of the visibility of the right upper and lower pulmonary vein. After extracting the cardiac blood masses the result gets triangulated and remeshed for activation time imaging. RESULTS: With this method the cardiac models of eight patients were extracted and the AT imaging approach was applied to single-beat ECG data of atrial and ventricular depolarization. CONCLUSION: The advantage of the proposed AAM approach is that only a few initial parameters have to be set. Therefore, the approach can be integrated into a processing pipeline that works semi-automatically. The extracted models can be used for further investigations.

Atrial and ventricular myocardium extraction using model-based techniques.

OBJECTIVES: This paper presents an efficient approach for extracting myocardial structures from given atrial and ventricular blood masses to enable non-invasive estimation of electrical excitation in human atria and ventricles. METHODS: Based on given segmented atrial and ventricular blood masses, the approach constructs the myocardial structure directly, in the case that the myocardium can be detected in the volume data, or by using mean model information, in the case that the myocardium cannot be seen in the volume data due to image modalities or artefacts. The approach employs mathematical and gray-value morphology operations. Regulated by the spatial visibility of the myocardial structure in the medical image data especially the atrial myocardium needs to be estimated repeatedly using the a-priori knowledge given by the anatomy. RESULTS: The approach was tested using eight patient data sets. The reconstruction process yielded satisfying results with respect to an efficient generation of a volume conductor model which is essential when trying to implement the estimation of electrical excitation in clinical application. CONCLUSION: The approach yields ventricular and atrial models that qualify for cardiac source imaging in a clinical setting.

A signal processing pipeline for noninvasive imaging of ventricular preexcitation.

OBJECTIVES: Noninvasive imaging of the cardiac activation sequence in humans could guide interventional curative treatment of cardiac arrhythmias by catheter ablation. Highly automated signal processing tools are desirable for clinical acceptance. The developed signal processing pipeline reduces user interactions to a minimum, which eases the operation by the staff in the catheter laboratory and increases the reproducibility of the results. METHODS: A previously described R-peak detector was modified for automatic detection of all possible targets (beats) using the information of all leads in the ECG map. A direct method was applied for signal classification. The algorithm was tuned for distinguishing beats with an adenosine induced AV-nodal block from baseline morphology in Wolff-Parkinson-White (WPW) patients. Furthermore, an automatic identification of the QRS-interval borders was implemented. RESULTS: The software was tested with data from eight patients having overt ventricular preexcitation. The R-peak detector captured all QRS-complexes with no false positive detection. The automatic classification was verified by demonstrating adenosine-induced prolongation of ventricular activation with statistical significance (p <0.001) in all patients. This also demonstrates the performance of the automatic detection of QRS-interval borders. Furthermore, all ectopic or paced beats were automatically separated from sinus rhythm. Computed activation maps are shown for one patient localizing the accessory pathway with an accuracy of 1 cm. CONCLUSIONS: The implemented signal processing pipeline is a powerful tool for selecting target beats for noninvasive activation imaging in WPW patients. It robustly identifies and classifies beats. The small beat to beat variations in the automatic QRS-interval detection indicate accurate identification of the time window of interest.

Computationally efficient noninvasive cardiac activation time imaging.

OBJECTIVE: The computer model-based computation of the cardiac activation sequence in humans has been recently subject of successful clinical validation. This method is of potential interest for guiding ablation therapy of arrhythmogenic substrates. However, computation times of almost an hour are unattractive in a clinical setting. Thus, the objective is the development of a method which performs the computation in a few minutes run time. METHODS: The computationally most expensive part is the product of the lead field matrix with a matrix containing the source pattern on the cardiac surface. The particular biophysical properties of both matrices are used for speeding up this operation by more than an order of magnitude. A conjugate gradient optimizer was developed using C++ for computing the activation map. RESULTS: The software was tested on synthetic and clinical data. The increase in speed with respect to the previously used Fortran 77 implementation was a factor of 30 at a comparable quality of the results. As an additional finding the coupled regularization strategy, originally introduced for saving computation time, also reduced the sensitivity of the method to the choice of the regularization parameter. CONCLUSIONS: As it was shown for data from a WPWpatient the developed software can deliver diagnostically valuable information at a much shorter span of time than current clinical routine methods. Its main application could be the localization of focal arrhythmogenic substrates.