Modelling phenomenological differences in aetiologically distinct visual hallucinations using deep neural networks.

Area of Science:

• Neuroscience
• Computational Psychiatry
• Psychology

Background:

• Visual hallucinations (VHs) are complex perceptions lacking external stimuli.
• Phenomenological differences exist between VHs arising from diverse etiologies like neurodegenerative diseases, visual loss (Charles Bonnet Syndrome), and psychedelic use.
• Understanding the mechanistic basis of these phenomenological variations is crucial.

Purpose of the Study:

• To investigate the mechanistic basis of phenomenological differences in VHs across distinct etiologies.
• To develop and validate a computational (neuro)phenomenology approach for modeling VHs.
• To generate synthetic VHs representative of specific etiological groups.

Main Methods:

• Utilized a coupled classifier and generative deep neural network to visualize learned representations.
• Examined three distinct etiological groups: neurodegenerative conditions, visual loss (CBS), and psychedelic users.
• Identified and manipulated three phenomenological dimensions: realism, spontaneity, and complexity to generate synthetic VHs.

Main Results:

• Generated synthetic VHs that captured the characteristic phenomenology of each etiological group.
• Experimental validation confirmed distinct phenomenological differences across the three dimensions between groups.
• Synthetic VHs were rated as representative of the actual hallucinatory experiences within each group.

Conclusions:

• Visual hallucinations exhibit significant phenomenological diversity linked to their underlying causes.
• Computational (neuro)phenomenology offers a powerful tool for modeling and understanding the characteristics of VHs.
• This approach successfully captures and differentiates the visual qualities of hallucinatory experiences across various conditions.