Patterns of functional damage in neural network models of associative memory

Area of Science:

• Computational neuroscience
• Network science
• Neurology

Background:

• Understanding neural network damage is vital for neurological and psychiatric disorders.
• Current knowledge of damage effects on neural networks is limited.
• Clinical observations in multi-infarct dementia highlight the importance of lesion characteristics beyond size.

Purpose of the Study:

• To present a simple analytical framework for estimating functional damage from focal structural lesions in neural network models.
• To quantify the effects of lesion area, shape, and number on associative memory retrieval.
• To derive scaling laws for lesion impact that can be experimentally tested.

Main Methods:

• Development of an analytical framework for neural network damage assessment.
• Quantification of lesion effects on a spatially organized associative memory model.
• Analysis of lesion properties including area, shape, and number.

Main Results:

• Multiple focal lesions impair performance more than a single lesion of equivalent size.
• Slit-like lesions are more detrimental to network function than rounder lesions.
• Larger networks exhibit less performance decrease for the same fraction of damage.

Conclusions:

• The number of lesions, rather than their total size, significantly impacts network performance.
• The framework explains the detrimental effect of infarct number in dementia and the interaction between Alzheimer's disease and multi-infarct dementia.
• Findings provide testable predictions for experimental neuroscience and clinical neurology.