Fig. 6
From: DiviK: divisive intelligent K-means for hands-free unsupervised clustering in big biological data
Cluster label normalisation procedure. Red = false negative; yellow = false positive; grey = true positive tumour. In the first step, a binary decision was made on whether the label should belong to one of the pathologist-defined regions. Clusters were sorted by the percentage of their area covered by the ROI. They were selected sequentially to optimise the Dice index. Secondly, all ambiguous assignments were resolved via optimising the Rand Index to form the normalised labels