Extended Data Fig. 3: Male examples of radiologist’s decision with and without AI assistance.
From: A foundation model for breast and lung cancer screening using non-contrast computed tomography
Each case was first read independently by radiologists, then re-read with AI-generated prediction score and heatmaps overlaid on CT images. Heatmaps were derived from the normalization layer in the final stage to highlight regions contributing to the prediction. Lung cancer cases which were (a) identified and (b) missed by OMAFound. (c) A non-lung cancer case which was overdiagnosed by OMAFound.
