Using fMRI to predict training effectiveness in visual scene analysis

Abstract

Visual analysis of complex real-world scenes (e.g. overhead imagery) is a skill essential to many professional domains. However, little is currently known about how this skill is formed and develops with experience. The present work adopts a neuroergonomic approach to uncover the underlying mechanisms associated with the acquisition of scene expertise, and establish neurobehavioral markers for the effectiveness of training in scene imagery analysis. We conducted an intensive six-session behavioral training study combined with multiple functional MRI scans using a large set of high-resolution color images of real-world scenes varying in their viewpoint (aerial/terrestrial) and naturalness (manmade/natural). Participants were trained to categorize the scenes at a specific-subordinate level (e.g. suspension bridge). Participants categorized the same stimuli for five sessions; the sixth session consisted of a novel set of scenes. Following training, participants categorized the scenes faster and more accurately, reflecting memory-based improvement. Learning also generalized to novel scene images, demonstrating learning transfer, a hallmark of perceptual expertise. Critically, brain activity in scene-selective cortex across all sessions significantly correlated with learning transfer effects. Moreover, baseline activity (pre-training) was highly predictive of subsequent perceptual performance. Whole-brain activity following training indicated changes to scene- and object-selective cortex, as well as posterior-parietal cortex, suggesting potential involvement of top-down visuospatial-attentional networks. We conclude that scene-selective activity can be used to predict enhancement in perceptual performance following training in scene categorization and ultimately be used to reveal the point when trainees transition to an expert-user level, reducing costs and enhancing existing training paradigms.