Loading
Accuracy is the share of predictions that were right; macro-F1 averages the F1 score of each class equally, so a model cannot look good by getting only the common cancer types right.
The F-score is a measure of predictive performance for binary classification calculated from precision and recall (Wikipedia). For a multi-class task such as cancer-type prediction, macro-F1 averages per-class F1 with equal weight, which exposes failures on rare classes that overall accuracy hides; a confusion matrix shows which classes are mixed up, often tissues of shared origin. Accuracy on a task saturated by an easy signal (tissue identity) leaves no room to show a new modality helps, a ceiling effect.
Shares ROC-AUC, PR-AUC and time-dependent AUC, Tissue-of-origin signal in tumour data, Cancer AI vocabulary (CanSim terms map) and the tag cansim-terms.
Shares Ablation study and multi-task heads, Cancer AI vocabulary (CanSim terms map) and the tag cansim-terms.
Shares Tissue-of-origin signal in tumour data, Cancer AI vocabulary (CanSim terms map) and the tag cansim-terms.
Shares Ablation study and multi-task heads, Cancer AI vocabulary (CanSim terms map) and the tag cansim-terms.
Shares Tissue-of-origin signal in tumour data, Cancer AI vocabulary (CanSim terms map) and the tag cansim-terms.
Shares Tissue-of-origin signal in tumour data, Cancer AI vocabulary (CanSim terms map) and the tag cansim-terms.
Shares ROC-AUC, PR-AUC and time-dependent AUC, Cancer AI vocabulary (CanSim terms map) and the tag cansim-terms.
Shares Cancer AI vocabulary (CanSim terms map) and the tag cansim-terms.