PT - JOURNAL ARTICLE AU - Sneha Shashidhara AU - Floortje S. Spronkers AU - Yaara Erez TI - Localizing the ‘multiple-demand’ frontoparietal network in individual subjects AID - 10.1101/661934 DP - 2019 Jan 01 TA - bioRxiv PG - 661934 4099 - http://biorxiv.org/content/early/2019/06/06/661934.short 4100 - http://biorxiv.org/content/early/2019/06/06/661934.full AB - The frontoparietal ‘multiple-demand’ (MD) control network plays a key role in goal-directed behavior. Recent developments of multivoxel pattern analysis (MVPA) for fMRI data allows for more fine-grained investigations into the functionality and properties of brain systems. In particular, in the MD network MVPA was used to gain better understanding of control processes such as attentional effects, adaptive coding, and representation of multiple task-relevant features. MVPA is often used with a region-of-interest (ROI)-based approach, in which distributed patterns of activity within ROIs are used to discriminate task-related neural representations. Common practice involves the use of an ROI template mask, ensuring that the same brain areas are studied for all participants. Another approach, commonly used in the visual system, is to define ROIs in individual subjects based on clusters of activity in an independent localizer task contrast. However, in the MD network the spread of activity is scattered and highly variable between participants, meaning that using a large template might not capture well the required areas in individual subjects, and clusters of activity may be difficult to define. To better localize this network at the individual level, we propose a hybrid conjunction masking approach, in which a group template of the network is used together with subject-specific independent localizer task data to select voxels with the highest levels of activity within the group template to be used in MVPA. To validate this approach, participants performed in the scanner three localizer tasks, spatial working memory, verbal working memory and a Stroop-like task, as well as a rule-based criterion task. We systematically assessed the three localizers based on their spatial spread of activity and levels of decoding in the criterion task when data from each localizer was used for voxel selection. A whole-brain analysis showed that all localizer tasks recruited the MD network at the group level, but individual patterns of activity were highly variable. Subsequent analysis of the extent of activation patterns, their specificity, and their consistency across runs revealed a similar picture of variable and distributed activity for the three localizers. Importantly, selecting voxels for MVPA based on the localizers’ data captured the underlying neural representations and yielded similar decoding results for all localizers as well as when all voxels within the group template were used. Together, these results suggest that all localizers may be suitable to identify the MD network in individual subjects. We propose that combining group level masks and individual subject data to localize the MD network allows for a refined and targeted localization while maintaining the decodability of task-related neural representations to further study the function and organization of the network.