355 Connecting computational models of reading to the brain in post-stroke alexia

Abstract

OBJECTIVES/GOALS: Many left hemisphere stroke survivors have a reading disorder (alexia), which is experienced as decreasing well-being. Therapies produce inconsistent results, demonstrating a need for treatment response predictors. We are investigating neural correlates of reading computational models to identify biomarkers to improve therapeutic outcomes. METHODS/STUDY POPULATION: Artificial neural network models of reading, mapping between orthography (visual word form), phonology (auditory word form), and semantics (word meaning), are trained to read single words at a healthy, adult capacity. The models are independently damaged to reflect the individual orthography-to-semantics, semantics-to-phonology, and orthography-to-phonology deficits of a sample of left hemisphere stroke survivors (n = 85). These deficits are measured with cognitive tests assessing the intactness of mappings between representations. Model damage is enacted by removing percentages of the connections between representations. For each type of deficit, the percentages of links removed are entered into a voxel-based lesion symptom mapping analysis to identify areas of cortex associated with that mapping. RESULTS/ANTICIPATED RESULTS: We anticipate that the neural correlates of model layers will be localized to a mostly left-lateralized network. Increased damage to the links between semantics and phonology in the model will likely be related to lesions involving the left posterior superior temporal sulcus and inferior frontal gyrus (IFG). Damaged orthography-to-semantic links will be related to the left fusiform gyrus (FG) and IFG. Finally, damage to the orthography-to-phonology links will be related to the left FG and superior temporal gyrus. DISCUSSION/SIGNIFICANCE: Mapping components of language models onto the brain will improve our understanding of the neural networks supporting language processing. Identifying these neural correlates may also produce biomarkers that can be used in predicting reading impairment at the acute stage or optimizing therapy in the chronic stage of stroke.