Evaluating the cognitive and electrophysiological plausibility of a thalamic computational model

The thalamus acts as a gateway to the cortex, relaying information from the brainstem and spinal cord to cortical regions. Evidence suggests that thalamic networks play a role in pattern recognition by extracting key elements from sensory input. We propose that thalamic networks function as a central orthogonalizer in the brain, enabling the application of the Hebbian learning rule without contamination or “interference” between the correct output for one input pattern and the output for other input patterns. This study aims to describe a biologically plausible artificial neural network that mimics aspects of the physiological activity of the thalamic circuit. To validate this proposal, the network was tested in four different scenarios. The model successfully replicated electrophysiological processes, including: (i) inhibitory facilitation; (ii) waveform sculpting in reticular cells; (iii) pattern completion of thalamic input; and (iv) computational processing of stabilized images on the retina. In human experiments, stabilized retinal images were perceived as sequences of patterns that appeared and disappeared suddenly. These patterns closely resembled or were related to the presented image, suggesting that they represent a mix of principal components derived from learned images. Such components emerge when a complete image, or only a portion of it, reaches the thalamus. Our model effectively reconstructs images based on partial eye-to-thalamus information, mirroring human visual responses.