A Three-Modules Scenario In An Interpretation Of Visual Hallucination In Dementia With Lewy Bodies And Preliminary Results Of Computer Experiments

This paper reports a three-modules scenario in a computational interpretation of Recurrent Complex Visual Hallucination in Dementia with Lewy Bodies (RCVH-DLB). An algorithmic representation of the interpretation is proposed and preliminary results of computer experiments based on a recurrent neural network model consisting of three modules are shown. The three modules are considered to be visual module, memory module, and perceiving module, respectively. Intra-module and inter-module interactions via asymmetric synaptic connection strength are introduced, where each module consists of a rather large number of binary state neurons. Corresponding to certain inputs to visual module, veridical perception is conceptualized as dynamic convergence to particular states (perceiving attractors) of neuron activity patterns in high dimensional state space. The key idea is that, as a possible origin of hallucination, systematic reducing of neurons and pruning of synaptic connections following the observed data in physical measurements of grey matter loss and white matter atrophy in DLB patients obtained by using DTI-MR, MRI or SPECT are introduced. Preliminary experimental results suggest that these defects can cause destabilizing of perceiving attractors and mutual penetration of them which results in hallucination-like outputs in perceiving module can occur.