Table 3 Results of the CNN on the five classes WSI classification task on the SKET mislabeled samples, considering both the models trained with automatically and manually labeled WSIs.

From: Unleashing the potential of digital pathology data by training computer-aided diagnosis models without human annotations

Dataset

Micro-accuracy (SKET labels)

Micro-accuracy (GT labels)

Weighted F1-score (SKET labels)

Weighted F1-score (GT labels)

Catania

0.817 ± 0.037

0.831 ± 0.022

0.579 ± 0.032

0.599 ± 0.039

Radboudumc

0.835 ± 0.008

0.851 ± 0.014

0.571 ± 0.027

0.627 ± 0.036

Back to article page