Fig. 8: Model consistency and robustness across experimental conditions. | Communications Medicine

Fig. 8: Model consistency and robustness across experimental conditions.

From: Volumetric spline-based Kolmogorov-Arnold architectures surpass CNNs, vision transformers, and graph networks for Parkinson’s disease detection

Fig. 8

a Coefficient of variation analysis showing AUC measurement stability across analyses. Blue bars represent 2D models, red bars represent 3D models, showing coefficient of variation values (left y-axis). Individual AUC values used to calculate each CV are overlaid as scatter points (right y-axis), demonstrating the underlying performance measurements across different analysis types. b Relationship between dataset sample size and performance variability. Blue circles represent 2D models, red circles represent 3D models, with dashed trend lines in corresponding colors. Dataset labels indicate PPMI (n = 59 subjects), NEUROCON (n = 43 subjects), and Tao Wu (n = 40 subjects). Error bars in a represent the standard error of the mean across analysis types.

Back to article page