Pairwise neural networks predict brain traits from limited MRI data

Inferring continuous cognitive traits from limited neuroimaging data using pairwise convolutional neural networks

Artificial IntelligenceComputer Vision and Pattern RecognitionMachine Learning

Summary

People's cognitive skills differ a lot, but it's hard to connect these differences to brain pictures using traditional methods, especially when you don't have many brain scans. The authors developed a new computer method that looks at pairs of brain images and compares them, instead of just looking at one at a time. This approach works well even with a small number of MRI brain scans and can predict how well someone performs on certain thinking tasks. It also helps identify which brain areas are important for those tasks.

What this means in practice

  • For neuroimaging analysts: Predict cognitive task performance using limited structural MRI data by comparing brain images in pairs with a convolutional model.
  • For clinical decision teams: Identify brain regions linked to cognitive abilities for personalized assessment when only small MRI datasets are available.

Authors

Marjan Rashidi, Robert Bain, Craig Stark, Jeffrey Krichmar, Mary Hegarty, Elizabeth Chrastil

Abstract

Individual differences in cognitive abilities vary widely, yet the structural brain features supporting these differences remain poorly understood. Traditional brain-behavior analyses rely on predefined metrics such as volume or cortical thickness and typically assume linear relationships with behavior, limiting sensitivity in small datasets. Here, we introduce a pairwise convolutional ranking framework that learns directly from whole-brain T1-weighted MRI by making comparative rather than absolute predictions, which performs well even with small samples (N=108). We tested this method with four small-scale spatial abilities tasks. Regression analysis predicted performance with high accuracy across all tasks, achieving greater predictive power over standard regression and pointwise deep learning baselines. Iterative cortical-region removal (backward elimination) revealed brain regions supporting individual performance common to all four tasks, as well as task-specific regions. These results demonstrate that pairwise learning extracts meaningful behavioral information from structural MRI, providing prediction and interpretable insight into the neuroanatomy of individual differences.