Fig. 5: performance of LCTfound on pixel level tasks.

a Differences in clinical processes for acquiring non-contrast lung CT images and lung CTA images. Lung CTA involves administering iodinated contrast agents intravenously and using CT for rapid imaging, requiring time for contrast agents to circulate in the body of patient. b Conventional lung CT scans generally necessitate reconstructing images from hundreds to thousands of views. Sparse-view CT imaging requires only a few angles, thereby substantially reducing the radiation exposure to patients. c Evaluation of virtual CTA results for LCTfound and three other methods. The metrics, from left to right, are sequentially: PSNR, SSIM, VIF, RMSE, CLIPIQA, LPIPS. For the first three metrics, their numerical values have a positive correlation with the quality of the image; for the latter three, the correlation is negative. d A case of virtual CTA results. Sequentially from left to right, the non-contrast lung CT image, the paired lung CTA image, the outcome of LCTfound, the outcome of CTA-GAN, the outcome of Pixel2pixel, and the outcome of CondDDPM. The red arrow indicates the virtual CTA imaging result of a tiny blood vessel. e Assessing of few-shot learning results for sparse-view reconstruction of lung CT. With 16 projection views, the pre-trained LCTfound obtained improved reconstruction outcomes. f Two cases of Lung CT for 16 projection views. The sequence from left to right includes the FBP result, the result from DIF-Gaussian, the result from SAX-NeRF, the result from DuDoTrans, the result from LCTfound, and the ground truth image. The images in the second row are enlargements of the areas within the yellow boxes in the first row of images. Fewer artifacts are evidently present at the site pointed by the yellow arrow in the LCTfound results. Statistical comparisons were performed using the two-sided Wilcoxon signed-rank test. Statistical significance is indicated as follows: ^*p < 0.05, ^**p < 0.01, ^***p < 0.001, ^****p < 0.0001; results without asterisks are not statistically significant (p ≥ 0.05). In the box plots, each point represents the score of an individual sample. The box indicates the interquartile range (IQR), with the middle line marking the median and the black dot representing the mean. Whiskers extend to 1.5 ×  IQR, and any points beyond this range are plotted as individual outliers.