Automatic diagnosis of imbalanced ophthalmic images using a cost-sensitive deep convolutional neural network

Table 2 The relevant parameters of conventional methods

Classifiers	Features extraction methods [27, 29]	Data-level methods [18, 19, 22]
Random forest (RF): number of trees = 300 Support vector machine (SVM): linear kernel function	Color and texture features (COTX): gray tone spatial dependence matrices: d = 1; gray gradient co-occurrence matrices: Lg = 10	SMOTE and BorSMOTE: nearest neighbors k = 5, the ratio of positive and negative samples r = 1 Under-sampling: the ratio of positive and negative samples r = 1
	Local binary pattern (LBP): P = 9
	Wavelet transformation (WAVE): two level wavelet transformation, Haar wavelet
	Scale-invariant feature transform (SIFT)

ISSN: 1475-925X