Table 1 Summary of ROC AUC and log loss results of the two training methods, fully-supervised (FS) and weakly-supervised (WS), on the 4 independent test sets: Kyushu Medical Centre, Mita Hospital, TCGA, and TCIA.
From: Weakly-supervised learning for lung carcinoma classification using deep learning
Kyushu Medical Centre | Mita Hospital | TCGA | TCIA | ||
|---|---|---|---|---|---|
WS | ROC AUC | 0.975 (0.950–0.993) | 0.974 (0.962–0.985) | 0.988 (0.981–0.994) | 0.981 (0.969–0.990) |
Log loss | 0.1338 | 0.4045 | 0.1819 | 0.3506 | |
FS | ROC AUC | 0.937 (0.908–0.958) | 0.922 (0.898–0.945) | 0.880 (0.840–0.917) | 0.963 (0.944–0.977) |
Log loss | 0.7022 | 0.8138 | 0.2977 | 0.2813 | |
