A COMPUTATIONAL MODEL FOR VISUAL CONFABULATION IN ANTON BABINSKI SYNDROME

Anton Babinski Syndrome (ABS) is a rare neurological condition characterised by cortical blindness, denial of visual impairment, and confabulation, with some improvement as patients gain awareness of their condition. The underlying pathophysiological mechanisms of ABS remain poorly understood due to its rarity and diverse clinical presentations, ranging from denial of blindness to complex confabulations. Here, we employ a feedforward visual processing model to elucidate plausible neurological bases for the clinical heterogeneity of anosognosia and the phenomenon of visual confabulation. We utilised a biologically inspired neural network, the HMAX model, to simulate feedforward visual processing and demonstrate the influence of top-down activity on pathophysiologies of ABS. Using the Caltech101 image dataset alongside a bespoke no-light category, we modelled visual experiences under varying lighting conditions. A linear kernel support vector machine was trained to classify these conditions within the HMAX feature space, allowing systematic exploration of neural encoding mechanisms. Our results indicate that ambiguous neural encoding may systematically bias against the recognition of visual deficits, supporting theories of self-generated activity in ABS. These findings suggest that top-down driven activity in surviving neural networks may effectively mask the lack of feedforward visual input, thereby contributing to the vivid confabulations and persistent denial of blindness seen in ABS. This computational framework provides a plausible mechanistic account for the variability in anosognosia presentations and offers insights into the mechanisms underlying progressive awareness of visual deficits. The study underscores the value of biologically-based computational models for understanding neurological disorders with heterogeneous and primarily qualitative clinical understandings.