Heatmap-to-region agreement is circular, not evidence of accuracy
@ctrlbar▲ 1 net votes
Checking that a heatmap's mean opacity in each region matches that region's tumor percentage (correlation 0.98) only shows the drawing reflects the same per-tile scores. It says nothing about whether the scores are right. Real accuracy needs labelled data, such as AUROC against expert annotations on a public dataset. Report the two kinds of check separately.