Jawshan Ara¹ (), Alireza Tavakkoli¹, Michael A. Crognale¹; ¹Institute of Neuroscience, University of Nevada, Reno, USA

Pattern VEPs can noninvasively quantify activity from specific achromatic and chromatic visual pathways (e.g., Murray et al., 1987; Kulikowski et al., 1989; Rabin et al., 1994). Achromatic pattern reversal (PR), which includes high spatial frequencies, is typically employed to stimulate the transient magnocellular pathway. Conversely, chromaticity modulation along equiluminant L-M cone opponent axis and the S-cone axis, using lower spatial frequencies and pattern-onset (PO) modes, is best suited to activate the more sustained chromatic pathways. VEP waveforms recorded with these parameters usually show stereotypical shapes. However, when stimulus parameters deviate from these optimal conditions, waveforms can vary significantly between individuals. For instance, achromatic PO-VEPs often exhibit greater inter-individual variability (e.g., Crognale et al., 1998; Crognale, 2002; Odom et al., 2016), since they reflect contributions from both magnocellular and chromatic pathways. Due to this variability, assessing the integrity and contributions of different neural pathways becomes challenging when stimulus parameters are not optimized as described. We classified VEPs using a Support Vector Machine (SVM) on VEP waveforms to classify the contribution of specific visual pathways. We also classified VEPs elicited by a broader, more general set of stimulus parameters. Next, we constructed a dataset to model VEP responses by varying the amplitude and latency of VEPs from visually normal subjects to simulate VEPs with visual pathologies. We then classified raw VEPs from visually normal subjects versus those from simulated visual pathologies, thereby establishing the clinical significance of simple machine learning algorithms for visual deficit detection. We plan to use SHAP (Shapley Additive exPlanations) analysis to rank VEP signal features, thereby enhancing the interpretability of the classification and extending these methods to multifocal VEPs to aid in the classification of localized visual deficits.