Fig. 5: The interpretability of OMAFound.
From: A foundation model for breast and lung cancer screening using non-contrast computed tomography
a, Heatmaps generated by five different post hoc interpretable approaches, including Grad-CAM, Grad-CAM++, Layer-CAM, Finer-CAM and attention-based GMAR. For breast cases, the right breast should match with the left side of the CT image due to anatomical opposite. b, Examples of non-cancer, lung cancer and breast cancer discrimination by OMAFound using preferable Finer-CAM, which were missed or partially missed by readers but well classified with the assistance of OMAFound. More examples including missed cases by OMAFound are shown in Extended Data Figs. 2 and 3.
