Automatic Quantification of Multiple Sclerosis Lesion Volume Using Stereotaxic Space
VBC '96 Proceedings of the 4th International Conference on Visualization in Biomedical Computing
An introduction to variable and feature selection
The Journal of Machine Learning Research
Learning to Decode Cognitive States from Brain Images
Machine Learning
Hi-index | 0.00 |
Background: Cortical features derived from magnetic resonance imaging (MRI) provide important information to account for human intelligence. Cortical thickness, surface area, sulcal depth, and mean curvature were considered to explain human intelligence. One region of interest (ROI) of a cortical structure consisting of thousands of vertices contained thousands of measurements, and typically, one mean value (first order moment), was used to represent a chosen ROI, which led to a potentially significant loss of information. Methods: We proposed a technological improvement to account for human intelligence in which a second moment (variance) in addition to the mean value was adopted to represent a chosen ROI, so that the loss of information would be less severe. Two computed moments for the chosen ROIs were analyzed with partial least squares regression (PLSR). Cortical features for 78 adults were measured and analyzed in conjunction with the full-scale intelligence quotient (FSIQ). Results: Our results showed that 45% of the variance of the FSIQ could be explained using the combination of four cortical features using two moments per chosen ROI. Our results showed improvement over using a mean value for each ROI, which explained 37% of the variance of FSIQ using the same set of cortical measurements. Discussion: Our results suggest that using additional second order moments is potentially better than using mean values of chosen ROIs for regression analysis to account for human intelligence.