Severity Classification of Conjunctival Hyperaemia by Deep Neural Network Ensembles

Table 3

Weighted kappa coefficients and kappa distance scores (KDS) for each model and expert graders.

Model	Κna	Κnb	Κab	Κnc	Κnd	Κcd	KDS
DenseNet201	0.693	0.707	0.748	0.712	0.643	0.653	−0.048
DenseNet121	0.693	0.713		0.704	0.642		−0.051
VGG19	0.692	0.708		0.714	0.637		−0.052
DenseNet169	0.656	0.690		0.718	0.663		−0.077
VGG16	0.679	0.692		0.681	0.651		−0.100
ResNet50	0.685	0.713		0.678	0.614		−0.114
InceptionV3	0.655	0.677		0.655	0.550		−0.266
Xception	0.580	0.669		0.619	0.538		−0.397
InceptionResNetV2	0.576	0.625		0.571	0.515		−0.517

Table is arranged in the order of the higher KDS in the evaluation of models. Κna: the weighted kappa coefficients of neural network and CO A. Κnb: the weighted kappa coefficients of neural network and CO B. Κab: the kappa coefficients of CO A and CO B. Κnc: the kappa coefficients of the neural network and CO D. Κnd: the kappa coefficients of the neural network and CO C. Κcd: the kappa coefficients of CO C and CO D. KDS: kappa scores.