Neural Connectivity Patterns Associated with High-Altitude Residence Duration: An Exploratory Cross-Validated Multimodal MRI Study.
Authors
Affiliations (7)
Affiliations (7)
- School of Psychology, Third Military Medical University, Chongqing, China.
- Department of Respiratory Disease, Xinqiao Hospital, Third Military Medical University, Chongqing, China.
- Department of Psychological and Cognitive Sciences, Tsinghua University, Beijing, China.
- Plateau Brain Science Research Centre, Tibet University, Lhasa, China.
- Department of Pediatrics, Fukang Hospital, Tibet University, Lhasa, China.
- Department of Psychological and Cognitive Sciences, Tsinghua University, Beijing, China. Electronic address: [email protected].
- Department of Psychological and Cognitive Sciences, Tsinghua University, Beijing, China. Electronic address: [email protected].
Abstract
Living at high altitudes (HA) poses challenges to the brain due to hypobaric hypoxia. However, most neuroimaging studies have focused on group-level comparisons, leaving the brain features associated with individual differences in HA residence duration incompletely characterized. We therefore used a data-driven, individualized, multimodal machine learning framework for exploratory cross-validated association analysis to estimate individuals' HA residence duration after adjustment for age and gender. The analysis included 217 healthy migrants at HA and incorporated whole-brain anatomical morphometry, structural connectivity (SC), and functional connectivity (FC). Resting-state FC showed the highest estimate-target correlation among the unimodal models (r = 0.34, permutation p = 0.01, out-of-sample predictive R² = -0.15, MAE = 41.04 months, RMSE = 52.39 months). Combining FC and SC produced a numerically higher r and lower MAE and RMSE (r = 0.36, permutation p = 0.01, out-of-sample predictive R² = 0.00, MAE = 38.37 months, RMSE = 48.86 months) than FC alone, but the difference in correlation was not statistically significant. The FC model thus showed a statistically significant but modest cross-validated association with the residualized target, but its accuracy was insufficient to support clinical prediction or monitoring of HA residence duration. Meanwhile, FC within the somatomotor network (SMN) showed a positive contribution to model estimation, while FC between the SMN and the default mode network (DMN) showed a negative contribution. Positive contributions were observed for SC between the DMN and the frontoparietal network and for FC between the DMN and the dorsal attention network. SC between the thalamus and the posterior superior temporal sulcus (pSTS) showed a positive contribution to model estimation. Whether these residence-duration-associated connectivity patterns reflect within-person changes over time requires longitudinal investigation. As no behavioral validation was performed in this study, the functional significance of these connectivity patterns remains to be determined.