Figure 4

Tumor margin estimation. A ground truth map was generated to highlight the tumor mass and margin (red), as well as the normal tissue (blue) and the boundary in between (black) (a). The tumor-normal boundary was superimposed on the H&E and the K-Means clustered image with PC2 data subset. The normal side mainly corresponds to cluster 1 and the tumor side is main associated with cluster 2 (b). Cluster occupancy barplot shows subset PC2 is the only data subset whose occupancy is reversed across the boundary (c). Difference between intra-cluster and inter-cluster occupancy across the boundary shows PC2 is best equipped to identify the tumor margin (d). Scale bar \(500\,\upmu\)m.