EFFECTIVE CONNECTIVITY IN RAT LATERAL GENICULATE NUCLEUS (LGN) ENHANCES DECODING OF VISUAL PATTERNS BEYOND SPIKE-BASED FEATURES

The lateral geniculate nucleus (LGN) has been regarded as a passive thalamic relay between the retina and primary visual cortex, but growing evidence suggests a higher-order role in visual information processing. To examine this, we recorded extracellular activity from the LGN of six adult rats using microelectrode arrays for approximately 26 minutes. The visual stimulus was a 4 × 8-pixel checkerboard in which four randomly selected pixels flickered ON for 200 ms, followed by 300 ms OFF; 32 distinct patterns were presented, each repeated 100 times. A cross-correlation analysis was performed to infer effective connectivity, and trial-specific adjacency matrices were used to train support vector machine classifiers to distinguish between all possible pattern pairs. Mean classification accuracy was compared for three representations of neural population: raw spike trains, spike counts, and connectivity matrices. For the top 10 stimulus pairs, accuracy was highest when using connectivity features (mean = 0.86), followed by spike counts (mean = 0.79) and spike trains (mean = 0.75). Effective connectivity outperformed the other representations in five out of six rats (Wilcoxon signed-rank test and t-test, p < 0.05), suggesting that network-based features enhance decoding performance. Tuning curves and receptive field analyses revealed that individual neurons did not consistently prefer the most accurately classified patterns, suggesting that improved decoding arose from emergent population-level dynamics captured through connectivity. These findings are consistent with a computational role for the LGN beyond simple signal relay. Our results underscore the value of connectivity-based representations that could inform future applications in visual prosthetics.