Fig. 7: Visualizations for model interpretability. | Nature Communications

Fig. 7: Visualizations for model interpretability.

From: General lightweight framework for vision foundation model supporting multi-task and multi-center medical image analysis

Fig. 7

a The heatmap shows the information acquired by the VFMGL for images in the MI and NMI classes. The red areas indicate a high level of model attention, while the blue areas indicate a low level of model attention. b The feature distribution of VFMGL in the four centers. c The score charts illustrate the MI and NMI cases of the four centers evaluated by the VFMGL. The statistical test used for this data analysis is the Independent t-test (two-tailed). In the training set, Center A (n = 410, p = 3.863e-22; mean ± std: 0.707 ± 0.109, 0.904 ± 0.128); Center B (n = 42, p = 0.014; mean ± std: 0.731 ± 0.078, 0.823 ± 0.093), Center C (n = 22, p = 0.265; mean ± std: -, 0.953 ± 0.017), Center D (n = 176, p = 9e-6; mean ± std: 0.846 ± 0.013, 0.943 ± 0.070); In the test set, Center A (n = 275, p = 1.249e-8; mean ± std: 0.753 ± 0.107, 0.875 ± 0.115); Center B (n = 29, p = 0.034; mean ± std: 0.721 ± 0.079, 0.815 ± 0.094), Center C (n = 15, p = 0.434; mean ± std: -, 0.948 ± 0.017), Center D (n = 118, p = 0.001; mean ± std: 0.858 ± 0.041, 0.940 ± 0.069). MI Myometrial Invasion, NMI Non Myometrial Invasion, PCA Principal Component Analysis, p significance value, std standard deviation. Source data are provided as a Source Data file.

Back to article page