2.3. Case study 3: continuous classification-based feedback

The participant first performed two fMRI runs to provide data to train an SVM classifier using the PRONTO toolbox [4]. A single fMRI run consisted of seven 20 s regulation blocks that were interleaved with seven 20 s baseline blocks. The participant was asked to perform visual-spatial imagery during the regulation blocks, and to look at the center of the screen during the baseline blocks. Next, the participant was asked to perform a similar real-time fMRI run with a similar design (10 regulation and 11 baseline blocks), where feedback was provided as expanding circle placed at the center of the screen. The feedback signal was computed using the dot product between the pre-trained classifier weight vector and the current data vector extracted from the classification mask [5], [6]. The experiment was performed at the Campus Biotech (Geneva) on a 3 T MR scanner (Prisma, Siemens Medical Solutions, Germany). Functional images were acquired with a single-shot gradient-echo T2*-weighted EPI sequence with 210 scans (32 channel receive head coil, TR=2020 ms, volume size=100×100×35 voxels, isotropic 2.2 mm³ voxel size, flip angle α=74°, bw=1565 Hz/pixel, TE=28 ms). The first 5 EPI volumes were discarded to account for T1 saturation effects.

For more details about the applied analyses, see the associated research article [1].