Predicting intelligence from fMRI data of the human brain in a few minutes of scan time

摘要: In recent years, the prediction of individual behaviour from fMRI-based functional connectome has become a major focus research. The motivation behind this research is to find generalizable neuromarkers cognitive functions. However, insufficient accuracies and long scan time requirements are still unsolved issues. Here we propose new machine learning algorithm for predicting intelligence scores healthy human subjects resting state (rsfMRI) or task-based fMRI (tfMRI). cohort 390 unrelated test Human Connectome Project, found correlations between observed predicted general more than 50~percent in tfMRI, around 59~percent when results two tasks combined. Surprisingly, that tfMRI data were significantly predictive rsfMRI even though they acquired at much shorter times (approximately 10~minutes versus 1~hour). Existing methods investigated benchmark comparison underperformed on produced well below our results. Our proposed differs existing it achieves dimensionality reduction via ensemble partial least squares regression rather brain parcellations ICA decompositions. addition, introduces Ricci-Forman curvature as novel type edge weight.