. 2019 Jul 4;1(1):20190021. doi: 10.1259/bjro.20190021

Table 1. .

The evaluation of the accuracy (A), interpretability (I) and explainability (E) of ML approaches in radiation outcomes prediction

Basic ML	Type	A	I	E	Improved ML	Type	A	I	E
Logistic regression^20,21	IP	*	****	***	GA²M⁶⁸	IP	**	***	**
					Ridge Regression²²	IP	**	**	*
					LASSO²³	IP	**	***	**
					Elastic Net^9,24	IP	***	**	*
Decision tree ^24,30,31	IP	**	*****	*****	CART³²	IP	***	****	*****
					Random Forests⁷	NIP	****	*	NA
					GBM^9,33	NIP	****	*	NA
					MediBoost^9,34	IP	****	**	*
Naïve BN ^35,37	IP	*	****	****	HBN^38,40	IP	**	***	**
Naïve BN ^35,37	IP	*	****	****	HBN-EK⁴¹	IP	**	****	***
Linear SVM ²⁴	NIP	**	**	*	SVM-RBF⁴³	NIP	***	*	NA
Linear SVM ²⁴	NIP	**	**	*	SVM-LRBF⁴⁴	NIP	***	**	*
Deep learning^49,50	NIP	****	*	NA	DL-HLV ^48,55,56	NIP	*****	**	NA
					DL-SA^52,57 /AM^59,60	NIP	*****	**	NA
					DL-DHLR^61–63	NIP	*****	***	NA
					DL-LIME⁶⁹	NIP	*****	***	NA

BN, Bayesian network; CART, classification and regression tree; DHLR, disentangled hidden layer representation; DL-AM, deep learning withattention mechanisms; DL-HLV, deep learning withcombination of handcrafted features and latent variables; GBM, gradient boosting machine; HBN, hierarchical Bayesian network; HBN-EK, hierarchical Bayesiannetwork with expert knowledge; HLV, handcrafted features and latent variables; IP, interpretable; LASSO, least absolute shrinkage and selection operator; LIME, local interpretable model-agnostic explanation; ML, machine learning; NIP, non-interpretable; SVM, support vector machine.