Text this: Statistical testing of agreement in overlap-based performance between an AI segmentation device and a multi-expert human panel without requiring a reference standard.