Josiane Mukahirwa¹, Qianli Meng, Keith Schneider; ¹University of Delaware, ²University of Delaware, ³University of Delaware

The magnocellular (M) pathway plays a critical role in both the visual and auditory systems, supporting the rapid processing of transient stimuli. Abnormal development of the magnocellular system has been proposed as a potential neural origin of developmental dyslexia, but remains debated, with alternative accounts emphasizing cortical malformations linked to language processing. One way to advance this debate is through machine-learning–based classification that evaluates the predictive power of thalamic versus cortical structures. Anatomical work by Galaburda & Livingstone (1993) demonstrated reduced magnocellular neuron size in the LGN and a hemispheric asymmetry in the MGN, with controls showing leftward dominance reversed in dyslexia. These findings suggest that thalamic magnocellular asymmetry may serve as a diagnostic marker, motivating fMRI approaches that model hemispheric differences. This study tests whether LGN/MGN asymmetries improve dyslexia classification. Nineteen participants (9 with dyslexia, 10 controls) underwent 3T fMRI during presentation of transient stimuli (abrupt onsets/offsets) and sustained stimuli (smooth temporal transitions). Regions of interest for the LGN and MGN were manually delineated on T1-weighted and proton-density images. Transient minus sustained (T–S) beta maps were computed. The MGN and LGN were each anatomically divided into five inferior-to-superior bins in the left and right hemispheres, generating feature sets designed to capture potential magnocellular–parvocellular gradients and hemispheric asymmetries. Feature selection was performed using LASSO regression with 500 bootstrap resamples; features selected in more than 70% of models were retained and entered into a logistic regression classifier. Classification performance was evaluated using leave-one-out cross-validation. The most inferior MGN bin in the left hemisphere, consistent with putative magnocellular layers, was the most consistently selected feature across bootstrap LASSO models. Logistic regression using this feature classified dyslexia with 78% accuracy. These findings provide functional evidence for thalamic hemispheric asymmetry in dyslexia, aligning with anatomical reports and supporting the subcortical magnocellular deficit hypothesis.