Table 1 ML algorithm and evaluation metrics per technical task are presented

Technical task	ML algorithm	Evaluation	Studies
Classification	Classic ML	ROC	^14,16,27
		Accuracy	^17,29
		Other (MCC, Precision, Recall, F1)	^17,30
	CNN	ROC	^15,32,36
	CNN	Other (Precision-Recall AUC, F1, Accuracy)	^32,36
	Large Pre-trained	ROC	^19,23,35
		F1 score	^24,38
		Accuracy	^{23,24,33,35,38}
		Other (Confusion matrices, Precision, Recall, Kappa coefficient)	^24,33,35,38
	GAN	ROC	^22,24
Detection	Large Pre-trained	ROC	^20,23,34
Detection	Other	Other (Precision, F1, MCC)	³⁴
Segmentation	UNet	Dice coefficient	^18,25,31,37
Segmentation	UNet	Other (IoU)	^18,37
Regression	Other (CNN)	Other (Dice coefficient, Accuracy)	^26,28

Publications occurring in multiple cells represent overlapping tasks and/or evaluation metrics. In papers where multiple ( > 1) tasks were involved, we only present the ML model for which evaluation metrics are reported. Note that we refer to technical tasks, ML algorithms and evaluation metrics, as reported in each publication. The term “Accuracy” presents a standalone metric (separate to ROC analysis), as reported by each paper authors. The term “Other” represents metrics that occurred once per ML model/technical task. ML: machine learning; ROC: receiver operating characteristic curve; MCC: Matthew’s correlation coefficient; AUC: area under the curve; IoU: intersection-over-union.

Quick links

Search