Figure 3
K-Means clustering of lung tissue. K-Means clustering was performed using all polarimetric and texture parameters for all sub-images, resulting in a binary cluster map (a). A silhouette score map of the K-Means results was generated to assess clustering performance (b). The binary and silhouette maps were combined into a single map of a continuous variable ranging from 1 (cluster 1) to 2 (cluster 2) using a mathematical transformation (c). Median filter of the continuous map provides for better visualization of the clusters (d). Performance of 8 different data subsets was assessed using silhouette scores of each cluster (e). The green dashed line indicates the largest silhouette score harmonic mean corresponding to PC1 and PC2. Median filtered maps of the data subsets highlight the 2 dimensional distribution of clusters, and their vertically averaged plots illustrate horizontal cluster variations (f–n). Scale bar \(500\,\upmu\)m.
