Fig. 2: Relevant imaging and clinical features selected via LASSO regression. | Nature Communications

Fig. 2: Relevant imaging and clinical features selected via LASSO regression.

From: Boundary complexity of cortical and subcortical areas predicts deep brain stimulation outcomes in Parkinson’s disease

Fig. 2

A Sixteen regional fractal dimension (FD) features and four clinical features were selected. The DBS target was one of the most important predictors of medication change following DBS, along with FD of three brain areas (B): the right paracentral lobule, left middle occipital gyrus, and right olfactory area. ROC AUC=area under receiver operating characteristic curve. C The LASSO shrinkage factor impacted the number of selected features and corresponding classification performance. The vertical red line denotes the optimal shrinkage factor used in the analysis. D Regional FD features selected via LASSO regression were spatially distributed throughout the brain. Warmer colors on the heat map denote features with higher ridge regression coefficients and thus greater importance for predicting DBS outcomes. E Comparison of reconstructed pial surface boundaries overlaid on T1-weighted images of two patients with high and low average FD, revealed visual differences in cortical structure. Source data are provided as a Source Data file.

Back to article page