Table 1.

Accuracy of tile classification performance of the four convolutional neural network (CNN) models on ballooning, inflammation, steatosis and fibrosis.

	Classification accuracy		N		Number of classes including “ignore”
	Train	Validation	Train	Validation
Ballooning	94.0%	93.1%	13590	1555	3
Inflammation	86.2%	86.0%	8567	914	4
Steatosis	93.8%	94.5%	6377	737	5
Fibrosis	88.5%	86.3%	4251	465	6

N is the number of labeled tiles used for training and validation.