Poster Presentation

Poster E in Poster Session E: Thursday, August 6, 10:30 am – 12:15 pm, Kimmel Center, Shorin & Rosenthal Rooms

Deep Graph Learning Reveals Individualized Somato-Cognitive Integration Zones in the Adolescent Brain: Evidence from the SCAN Network

Yuxin Xue¹, Wen Li², Jianyu Li³, Xin Jing¹, Jingguo Dai⁴, Haozheng Shen⁴, Luqi Cheng⁵, Weiyang Shi², Yuanchao Zhang¹, Yongfu Hao⁴, Huixiong Zhang¹, Yu Zhang⁶; ¹University of Electronic Science and Technology of China, ²Institute of Automation, Chinese Academy of Sciences, ³West China Hospital of Sichuan University, ⁴Zhejiang Lab, ⁵Guilin University of Electronic Technology, ⁶Shanghai Jiaotong University

Presenter: Yu Zhang

The discovery of the Somato-Cognitive Action Network (SCAN) challenges the classical motor homunculus by revealing inter-effector regions that integrate bodily control with executive functions. However, mapping these fine-grained integration zones in adolescents remains a computational challenge due to high inter-individual variability and the blurring effects of group-averaged templates. Here, we propose an uncertainty-aware Graph Neural Network (GNN) framework to delineate individualized functional architectures. Leveraging high-resolution fMRI (resting-state and movie-watching) from the Healthy Brain Network (HBN), our model incorporates cortical mesh topology as spatial constraints to generate vertex-wise areal labels, uncertainty maps, and integration scores. The GNN-derived individualized atlas demonstrated superior intra-subject homogeneity compared to group-level templates, successfully identifying individualized SCAN networks interdigitated between classical effector-specific areas (foot, hand, and tongue). Crucially, we revealed systematic differences in functional organization along the dorso-ventral axis common to both hemispheres: transitioning from dorsal planning zones (motor imagery/anticipation) to middle execution hubs (visuomotor coordination), and finally to ventral monitoring zones linked to error detection and linguistic processing. We further delineated ventral integration loops connecting SCAN to auditory cortices, establishing a perception-action cycle that transforms sensory inputs into motor plans. Finally, developmental analysis demonstrated that naturalistic stimuli enhance the detection of developmental trajectories. We observed age-related increases in SCAN hub participation, specifically stronger coupling between SCAN and Ventral Attention Networks in young adults compared to preadolescents. Moreover, we observed state-invariant age-related increases in participation coefficients across SCAN and Auditory-Linguistic hubs. These findings provide computational evidence that the SCAN serves as a dynamic, hierarchically organized mind-body interface that gradually matures to support complex real-world interactions, offering new targets for understanding neurodevelopmental conditions involving motor-cognitive disconnects.

Topic Area: Development, Individual Differences & Clinical Populations