gms | German Medical Science

The typical workflow of fMRI experiments separates data acquisition and data analysis. In contrast, real-time fMRI (rfMRI) combines these steps within one process and allows to provide feedback of the actual own mental state to the volunteers which enables task training and optimization of individual strategies to increase the BOLD response [Ref. 1], [Ref. 2]. Based on rfMRI volunteers can interact with the dynamical stimulus environment which may change depending on localization and intensity of activation. However, many real world actions involve communication between several partners (such as in economics), where highly non-linear interactions can usually not been foreseen, and therefore not be planed in advance in static environments.

We extended our flexible XML-controlled rfMRI software system to connect two (or more) MR scanners in real-time, and to feed back the activation parameters of both volunteers to each of them. Additionally the system provides the capability to change virtual reality paradigms depending on the activation patterns of both volunteers. We tested the setup with a competition paradigm thereby connecting a 3T and a 7T scanner. Compared to other approaches such as NEMO [Ref. 3] our system integrates the control of the scanners as well as the presentation and data analysis, enabling to adapt the paradigm according to the activation patterns e.g. to increase attention or to modify a virtual reality environment [Ref. 4].

Stimulus presentation, data analysis, and communication with each scanner system were realized using a custom-made MATLAB-based real-time fMRI software system [Ref. 5]. The software was extended to allow for central control of the data exchange between two MRI scanners. Experiments were conducted using a 3T and 7T MRI scanner (Siemens, Erlangen, Germany). Timing of stimulus presentation and communication of both scanner systems were defined in a central XML file. We used a modified EPI-BOLD sequence to allow for real-time export of the functional images. The software analyzes the data, extracts activation patterns and sends the results of the online data classification [Ref. 4], [Ref. 5], [Ref. 6] and the actual parameters such as the BOLD signal to a central SSH server accessible by both scanner environments. Parallel rfMRI was performed with four healthy volunteers (m, 22–28 years, right handed) after giving written consent according to the local ethics committee. The volunteers were introduced to each other and were subsequently instructed about the experiment setup. Measurement parameters were: TR 2000 ms, TE 21 ms (7T)/ 29 ms (3T), matrix size 64 x 64, 31 slices (3T)/ 7 slices (7T), resolution of 3.4 x 3.4 x 3 mm³ (3T) / 3.3 x 3.3 x 4.6 mm³ (7T), flip angle 90°. Motor tasks (finger tapping, both paced by spoken words “start”, “stop”) were used for the experiment which consisted of two parts: (1) an initialization measurement where the individual activation patterns for the defined task were semi-automatically identified; and (2) the parallel rfMRI experiment. Identification of individual activation patterns consists of histogram based thresholding, region connectivity check, region labeling and user selection of the target region(s).

A block design (4 sec baseline – 6 sec active – 14 sec baseline) was used for both tasks (5 runs for the initialization measurement, 60 runs for the parallel rfMRI session). Each volunteer was told to maximize the BOLD response in the previously determined brain areas visualized by a bar and a virtual sphere, and to reach higher activation scores than the volunteer in the other MRI scanner. During each single run we used the BOLD signal averaged over the previously identified activated regions as a measure for the activation level. It was (1) presented directly to the volunteer as a coloured bar, and (2) served for the progression of the volunteers virtual sphere (the larger the activation, the larger the progression). At the same time each volunteer saw the sphere of the opponent. To account for 3T vs. 7T signal differences both signals were scaled to 0 to 1 based on the BOLD responses of the initialization measurements.

To increase the subjects’ motivation their final financial reward was related to the achieved activation levels. Only the winner of the each single run is rewarded. Financial rewards, positions of the spheres, and colour of the bar were updated after each run (24 secs) in this experiment but the software allows also for continuous updates.

The described experiments were successfully conducted. The connection between two MRI scanners could successfully be established. The transfer of the results of the online data analysis, classification of activation patterns, and extraction of the BOLD response via SSH took below 100 ms. Because of 2000 ms TR and the delay of the BOLD response the transfer solution seems suitable for the exchange of online determined parameters between the involved scanner environments. As expected the increase of the financial reward as well as the competition with a second volunteer kept the motivation on a high level so that the level of BOLD responses became more stable over all runs during each experiment. The synchronization of both scanners and their analysis environments raised no problems although two firewalls had to be tunnelled. The result visualization was updated correctly after each run with the latest activation scores. Nevertheless the sequences have to be started simultaneously to enable synchronous scanning with shorter repetition times. The presented system is ideally suited to perform neuro-economic experiments.