Skip to main content
Fig. 1 | Breast Cancer Research

Fig. 1

From: Machine learning prediction of HER2-low expression in breast cancers based on hematoxylin–eosin-stained slides

Fig. 1

Overview of the study. a The patient cohort of the in-house USTC-BC dataset, which displays the proportions of different HER2 expression levels and the division of data for experiments. b The WSI pre-processing. Each WSI is divided into a set of patches and corresponding coordinates. c Overview of the AI model. The network structure of the AI model is mainly composed of the Kernel Attention Transformer (KAT) modules. The input of the AI model consists of two parts: the patch features extracted by a pre-trained feature extractor and the anchors generated by the coordinates of the patches. The output includes two parts: the probability corresponding to each predicted category and the attention scores used for interpretative analysis. d The structure of the KAT module. Each KAT module has two layer-normalization layers, one kernel-attention layer, and one feed-forward layer

Back to article page

Breast Cancer Research

ISSN: 1465-542X

Contact us