Skip to main content
. 2020 Nov 26;11:6004. doi: 10.1038/s41467-020-19817-3

Fig. 1. Strategies for the establishment of AI models for analysis of pathologic images.

Fig. 1

a Workflow of the system. Photographs of the slide images and scanned images of the entire slides for DLBCL and non-DLBCL were used for establishing AI models. b Overview of data flow implemented and components of the GOTDP-MP-CNNs. 17 CNNs [AlexNet, GoogleNet(ImageNet), GoogleNet(Places365), ResNet18, ResNet50, ResNet101, Vgg16, Vgg19, Inceptionv3, InceptionResNetv2, SqueezeNet, DenseNet201, MobileNetv2, ShuffleNet, Xception, NasNetmobile, Nasnetlarge] were utilized as a whole in generation of our AI models, and the transfer learning was used to optimize the fitness of the data. A specific AI model was established by training 80% of total samples with 10% of them used for model validation and the remaining 10% of the samples for testing diagnostic accuracy of the established model.