Skip to main content
. Author manuscript; available in PMC: 2022 Sep 17.
Published in final edited form as: Neurocomputing (Amst). 2021 Jan 23;453:312–325. doi: 10.1016/j.neucom.2020.04.153

Figure 4:

Figure 4:

The receiver operating characteristic (ROC) curves of thyroid slides diagnosis on pooling, self-attention, and concatenation fusion manners using the VGG16bn model as the unit feature extractor. From top to bottom are the ROC curves of thyroid’s three diagnosis categories, including Benign, Uncertain, and Malignant. The self-attention fusion attains higher AUC values compared to other two fusion manners.