Berkeley K Fahrenthold¹, Matthew R Cavanaugh¹, Michael D Melnick¹, Duje Tadin¹, Marisa Carrasco², Krystel R Huxlin¹; ¹University of Rochester, ²New York University

Cortical blindness (CB) arising from damage to early visual cortex can be partially alleviated by visual retraining, although perceptual thresholds at trained blind-field (BF) locations typically remain impaired relative to corresponding intact-field (IF) locations. Previous results suggest that high internal additive noise underlies residual BF impairments. In normally-sighted individuals, feature-based attention (FBA) increases gain and sharpens neuronal population tuning, helping overcome internal noise and reducing the impact of external noise. Here, we asked whether training with FBA engages similar mechanisms in the BF and IF of CB patients. Twenty-nine patients (54±12 years old, 19 males) enrolled 1.2–5.1 months post-occipital stroke (clinical trial NCT04798924) trained on a direction discrimination task with FBA across two phases. In Phase1, they were randomized to train in either their BF or IF until 6 months post-stroke; in Phase2, all participants trained only in their BF for the next 6 months. Direction discrimination thresholds improved significantly in the BF after both training phases (Phase1 improvement by 36.14±34.19˚; Phase2 improvement by 32.9±40.33˚) but only after Phase1 in the IF (Phase1 improvement by 0.77±1.18˚; Phase2 worsening by 0.13±0.81˚). Threshold versus noise (TvN) functions were measured pre- and post-training and fit using the perceptual template model (PTM) framework. This revealed key differences between the BF and IF: in Phase1, TvN curves in the IF showed improvements consistent with a combination of enhanced external-noise exclusion and reduced internal additive noise, with no further changes in Phase2. In the BF, TvN curves exhibited their largest improvements in Phase2, driven primarily by reductions in internal additive noise. Our results suggest that training with FBA elicits perceptual learning in the BF via different processing changes than in the IF. Whereas training in the IF changes both gain and tuning, training in the BF primarily reduces internal additive noise and increases population-response gain.