Fig. 1: Neuroanatomical feature models and representational similarity analysis. | Communications Biology

Fig. 1: Neuroanatomical feature models and representational similarity analysis.

From: Contrastive machine learning reveals Parkinson’s disease specific features associated with disease severity and progression

Fig. 1

a The VAE and CVAE models were employed to extract neuroanatomical features, and the resulting features were then used to construct dissimilarity matrices. b Comparisons of the similarity of the extracted neuroanatomical features with various nonclinical and clinical characteristics, the number of each independent characteristic please see Supplementary Table 1. To evaluate the reliability of RSA analyses across diverse samples obtained from the latent distributions of the CVAE, 10 samplings were conducted for each patient with PD. As a result, 10 unique reconstruction datasets were generated, and RSA models were fitted to each reconstruction separately. *p < 0.05; **p < 0.01; ***p < 0.005.

Back to article page