Fig 7. Illustration of the pixel-based metrics.

(a) True Positives (TP, nuclear pixels segmented as nucleus), true negatives (TN, background pixels segmented as background), false positives (FP, background pixels segmented as nucleus) and false negatives (FN, nuclear pixels segmented as background). These quantities were used to compute accuracy and Jaccard similarity index for the image-processing algorithm and three pre-trained deep learning architectures. (b) Segmentation overlaid on the same slice as (a) (slice 184/300—on a filtered image): ground truth (GT) in cyan, automated segmentation in purple.