Linear Neuroimaging Models

The study's conclusions are directly established for L2-regularized logistic regression rather than nonlinear models. Under collinearity, linear-model weights may reflect shared variance rather than region-specific effects. Correlated input features weaken the clinical interpretability of linear-model coefficients. Logistic regression was chosen because it is linear and interpretable. Linear models are widely used in computational neuroimaging because their coefficients can link brain measurements to clinical outcomes.