Using machine learning-based lesion behavior mapping to identify anatomical networks of cognitive dysfunction:...

Area of Science:

• Neuroscience
• Cognitive Neurology
• Neuropsychology

Background:

• Previous studies on spatial neglect after right brain damage used univariate methods with inconsistent results.
• This led to the hypothesis of a widespread network underlying spatial orientation, potentially undetectable by univariate approaches due to limited statistical power.

Purpose of the Study:

• To directly validate the network hypothesis of spatial neglect using a large patient sample.
• To compare univariate and multivariate lesion-mapping techniques for identifying the neural correlates of spatial neglect.

Main Methods:

• A large sample of 203 newly recruited right brain damaged patients was analyzed.
• Multivariate lesion-mapping (support vector regression) was compared with univariate analyses.
• Standardized correction factors (FDR correction) and lesion size control (dTLVC) were applied to both methods.

Main Results:

• Both univariate and multivariate approaches identified a similar complex network pattern associated with spatial neglect.
• Cortical lesions predominantly affected the temporal cortex, extending to inferior parietal and occipital areas.
• Affected regions also included frontal and subcortical gray matter and connecting white matter tracts.

Conclusions:

• The findings support the importance of a right-hemisphere network for spatial exploration and attention.
• Lesions within this network are critical for the emergence of core symptoms of spatial neglect.
• Both univariate and multivariate methods can detect this network when appropriately applied.