Table 3 ROC-AUC, log loss, accuracy, sensitivity, and specificity results for aggressive and indolent classification on the core needle biopsy test set using transfer learning (TL) and weakly supervised learning (WS) model (TL-Colon poorly ADC (x20, 512) and WS) and fully and weakly supervised learning model (TL-Colon poorly ADC (x20, 512) and FS+WS)
From: Inference of core needle biopsy whole slide images requiring definitive therapy for prostate cancer
| Â | aggressive | indolent |
|---|---|---|
TL-Colon poorly ADC (x20, 512) and WS | ||
   ROC-AUC | 0.970 [0.957 - 0.981] | 0.851 [0.819 - 0.885] |
   Log-Loss | 0.410 [0.320 - 0.500] | 1.133 [0.959 - 1.298] |
   Accuracy | 0.918 [0.898 - 0.940] | 0.758 [0.727 - 0.792] |
   Sensitivity | 0.885 [0.846 - 0.920] | 0.870 [0.798 - 0.933] |
   Specificity | 0.946 [0.925 - 0.968] | 0.738 [0.705 - 0.777] |
TL-Colon poorly ADC (x20, 512) and FS+WS | ||
   ROC-AUC | 0.980 [0.969 - 0.990] | 0.846 [0.813 - 0.879] |
   Log-Loss | 0.213 [0.160 - 0.260] | 2.273 [2.012 - 2.475] |
   Accuracy | 0.935 [0.918 - 0.957] | 0.736 [0.707 - 0.772] |
   Sensitivity | 0.946 [0.919 - 0.973] | 0.900 [0.833 - 0.955] |
   Specificity | 0.926 [0.899 - 0.955] | 0.706 [0.673 - 0.750] |