Skip to main content

Table 5 DICE Scores evaluating predictions from UNet, ColorAE, and ensemble methods against per-pixel (hand-drawn) labels

From: Deep learning-based image analysis methods for brightfield-acquired multiplex immunohistochemistry images

Model CD3 CD4 CD8 CD20 CD16 K17
ColorAE 0.649 0.543 0.764 0.589 0.539 0.649
UNet 0.649 0.463 0.676 0.554 0.581 0.649
Intersection 0.457 0.457 0.769 0.539 0.530 0.457
Union 0.553 0.540 0.675 0.599 0.591 0.553
Union anchor AE 0.611 0.548 0.685 0.588 0.593 0.611
Union anchor UNet 0.526 0.509 0.676 0.558 0.592 0.526