Fig. 3
From: An unsupervised XAI framework for dementia detection with context enrichment
Cluster Map. Hierarchical clustering dendrogram (on the Y-axis) resulting from Ward’s hierarchical clustering analysis of individual w-scores profiles of participants computed from three features - CNN relevances, volume, and cortical thickness measures. Four disease diagnoses were considered: cognitively normal (CN, color-coded as blue), mild cognitive impairment (MCI, color-coded as pink), dementia due to Alzheimer’s disease (AD, color-coded as red), and frontotemporal dementia (FTD, color-coded as green). The pie charts visualize the relative homogeneity, with respect to the disease diagnoses, of the two clusters. The W-score features are visualized using a custom color scale to indicate the extent of the deviation, where a gradual intensification of color (either red or green) signifies increasing pathological observations. For a vector graphic rendering, please refer to the GitHub version of the plot.
