Deep multiview learning to identify imaging-driven subtypes in mild cognitive impairment

BMC Bioinformatics

Table 1 Cluster evaluation

Clusters are evaluated using Calinski and Harabasz (CH) and the silhouette score as internal measures. Higher CH score and Silhouette close to 1 indicates more dense and well separated clusters, where lower CH score and Silhouette closer to 0 indicates more poorly defined and overlapping clusters. In addition, Adjusted Mutual Information (AMI) are calculated for each cluster assignment against the original EMCI/LMCI diagnosis. AMI closer to 1 means two sets of clusters are more similar where AMI close to 0 means they are more independent from each other

ISSN: 1471-2105