Nuclear shape, architecture and orientation features from H&E images are able to predict recurrence in node-negative gastric adenocarcinoma

Table 3 Evaluation of different combinations for feature selection and classifier validation on training set and test set

Dataset	Classifier	Feature selection	AUC	Accuracy	Specificity	Sensitivity
D1	ALD	WRST	0.77 ± 0.08	0.81 ± 0.08	0.82 ± 0.05	0.67 ± 0.03
		MRMR	0.67 ± 0.05	0.80 ± 0.04	0.84 ± 0.02	0.69 ± 0.09
		RF	0.76 ± 0.03	0.79 ± 0.06	0.85 ± 0.02	0.62 ± 0.05
	AQD	WRST	0.81 ± 0.02	0.74 ± 0.09	0.80 ± 0.03	0.71 ± 0.08
		MRMR	0.79 ± 0.06	0.77 ± 0.01	0.82 ± 0.01	0.73 ± 0.01
		RF	0.72 ± 0.06	0.87 ± 0.02	0.86 ± 0.03	0.75 ± 0.03
	RF	WRST	0.83 ± 0.03	0.79 ± 0.06	0.82 ± 0.06	0.72 ± 0.04
		MRMR	0.81 ± 0.06	0.76 ± 0.04	0.80 ± 0.08	0.70 ± 0.06
		RF	0.80 ± 0.05	0.73 ± 0.08	0.79 ± 0.06	0.69 ± 0.08
	SVM	WRST	0.87 ± 0.03	0.89 ± 0.02	0.88 ± 0.01	0.78 ± 0.08
		MRMR	0.84 ± 0.02	0.88 ± 0.01	0.84 ± 0.02	0.72 ± 0.04
		RF	0.81 ± 0.05	0.84 ± 0.04	0.82 ± 0.02	0.73 ± 0.07
D2	SVM	WRST	0.76	0.72	0.74	0.68

Evaluation values in italic indicate the best machine learning combination
MRMR minimum redundancy maximum relevance, RF random forest, WRST Wilcoxon rank sum test, LDA analysis of linear discriminant, AQD analysis of quadratic discriminant, SVM machine of support vector, AUC area under receiver operating curve

ISSN: 1479-5876