Skip to content

ANS and RNS metrics

ANS and RNS were introduced by Wang et al. (2024) as reconstruction-derived measures of hemispheric specificity.

Metric definitions and software interpretation

For the actual target-hemisphere GM value Act_i and its reconstructed value Recon_i, HemiSpec uses the following interpretation of the definitions introduced by Wang et al. (2024):

ANS_i = abs(Act_i - Recon_i)
RNS_i = abs((Act_i - Recon_i) / (Act_i + Recon_i))
  • ANS: absolute neuroanatomical specificity.
  • RNS: relative neuroanatomical specificity.

ANS represents the amount of actual–reconstructed difference. RNS represents the proportion of local signal expressed by that difference.

HemiSpec numerical implementation

HemiSpec computes:

ANS = abs(GM - recon)
RNS = abs(GM - recon) / (abs(GM) + abs(recon) + eps)

The implementation also requires finite values and applies a configurable valid-GM threshold (0.15 by default). The absolute-value denominator and small eps stabilize division when reconstructed values are near zero or contain small negative numerical values. For non-negative GM and reconstruction values, the implementation approaches the published expression as eps approaches zero.

This implementation detail should be reported when exact numerical reproducibility matters.

Interpretation limits

ANS/RNS are reconstruction-error-derived quantities. They are not direct measurements of neuronal function, causal lateralization, or tissue pathology. Interpretation depends on preprocessing quality, model domain, reconstruction behavior, masking, site/scanner effects, and downstream validation.

Citation boundary

Cite Wang et al. (2024) for the metric framework. Cite HemiSpec separately for the software implementation when a public software record is available. See Citation.