Abstract
Individual differences in domain-general object recognition abilities, as captured by the latent construct o, predict performance on several visual processing tasks. We advance knowledge in this area by investigating the relationship between o and category learning. Category learning can be supported by exemplar learning, which would be expected to relate to o, but also by rule abstraction, which may not. We investigate this relationship in the context of medical imaging, using a task in which novice participants must categorize white blood cell images as cancerous or not. In study 1, performance was reliable in a blood cell category learning task which provided feedback for only the first 10 of 60 trials (ω = .91). However, evidence negated a relationship between an aggregate measure of o (based on two tasks with novel objects) and accuracy in category learning (n = 77, r = .085, BF+0 = 0.28). In study 2, we randomly assigned participants to conditions with feedback on only the first 10 trials or on all trials of the blood cell category learning task. We found Bayesian evidence for a correlation between o and categorization accuracy when participants receive continuous feedback (n = 122, r = .269, BF+0 = 18.75), but not when they receive only 10 trials of feedback (n = 122, r = .11, BF+0 = 0.41). Moderate Bayesian evidence supports a difference between these correlations (BF+0 = 8.93). Supervised category learning is predicted by domain-general object recognition ability with an effect size similar to that previously found for learning to detect tumors in chest radiographs. Variability in mostly unsupervised category learning may instead be related to the effectiveness of abstract rules formed early in learning. Domain-general visual ability could complement other sources of individual differences when predicting the learning of accurate medical image interpretation.