Reject the shortcut: correlation describes the relationship between the axes; a useful label-oriented view also depends on complementary variation and the observed class overlap.

Accept it only when both features have larger values in Malignant samples, because that establishes a causal link.

Replace it with the opposite rule: the least correlated pair always gives the clearest diagnosis separation.

Accept it: a large feature-feature correlation is equivalent to a large separation between diagnosis groups.