Matt Cavanaugh¹, Berkeley Fahrenthold¹, Krystel Huxlin¹; ¹University of Rochester

A limitation of rehabilitation for stroke-induced cortical blindness (CB) is the burdensome nature of training interventions, which require hours of focused effort over weeks of daily training - a difficult task when adjusting to life post-stroke. Statistical learning, in which patterns of information are learned automatically, offers a potentially easier method of improving blind field performance without the concerted effort of traditional training. Here, nine CB participants and six visually-intact controls performed a single session of an 8AFC motion discrimination statistical learning task. In CB patients, motion discrimination was pre-trained to recovery at the blind field location targeted. In the statistical learning task, motion directions of a high-contrast stimulus were grouped into triplets, each repeated 21 times (high incidence) or 7 times (low incidence), with participants unaware of these groupings. CB participants performed above chance (8AFC chance=12.5% correct) on the statistical learning task, averaging 25.0+/-13.3% correct (p=0.02), while controls averaged 94.8+/-4.8%, with no difference seen for high- versus low-incidence trials in either cohort. To assess learning, triplets were presented in a two-interval forced-choice task, with participants indicating the more familiar triplet. Intervals contained high- and low-incidence triplets, and five “Null” triplets never shown during the learning phase. Two Null sets followed the design rules of regular triplets (Lawful), while three broke the rules (Unlawful). CBs averaged 48.4+/-5.7% correct on the familiarity task, while controls averaged 61.1+/-10.0% (p=0.021). However, significant differences emerged between Null cases, with controls (p=0.006) and CBs (p=0.02) performing better when distinguishing presented triplets from Unlawful Nulls versus Lawful Nulls. Thus, in a single session, CBs learned statistical rules within their blind field, despite poorer discrimination performance than controls. These results provide the first evidence that statistical learning is possible within CB fields, motivating rule-based learning as a tool to accelerate patient training, making rehabilitation less burdensome.