DECODING OF FACIAL EXPRESSIONS: EMOTIONAL VALENCE, CORTICAL TOPOGRAPHY, AND MOTOR INTEGRITY

Objective: To characterize facial expressions using Deep Learning and Optogenetics to integrate and dissect the neurobiological flow of gustatory valence encoding, the functional organization of the premotor cortex, and the integrity of the peripheral motor pathway.
Methods: Three strategies were employed in C57BL/6 and Thy1-ChR2-YFP mice. Three protocols were performed:
1. Optogenetic mapping of the Anterolateral Motor Cortex (ALM).
2. Facial expression analysis during gustatory stimulation (30-35 μL) with sucrose (20%), denatonium (1 mM), and water.
3. Evaluation of facial paralysis (compression vs. transection) during spontaneous locomotion.
Encoding was performed using: DeepLabCut (kinematics), PCA (trajectories), FaceMap (SVD), and HOGs (prototype/classification).
Results: Stimulation in Thy1 mice showed significant motor responses compared to baseline (Wilcoxon, p<0.05), revealing a functional topography in ALM with zone-dependent latencies (1.03s nasal vs 4s lateral). In C57BL/6, the tools discriminate
d against valence: HOGs differentiated the pleasure pattern (sucrose) from the aversion pattern (ANOVA, p<0.001; Tukey, p<0.001). Kinematics (DeepLabCut) demonstrated that said valence manifests through specific vertical changes in mouth and eyes (Kruskal-Wallis p<0.01), a finding corroborated by PCA, where the second component (PC2) exclusively isolated valence (p<0.01). Finally, under paralysis conditions, automated quantification detected significant alterations in whisker kinematics (p<0.05 vs control), validating the model's sensitivity to motor deficits.
Conclusions: The analysis demonstrates that facial expression is not random but follows an ordered flow, integrating valence encoding and the topographic organization of the premotor cortex, converging into a stereotyped kinematic execution that depends on the integrity of the peripheral motor pathway.