Synthesize Mammogram from Digital Breast Tomosynthesis with Gradient Guided cGANs
Compared to mammographic screening, digital breast tomosynthesis (DBT) as an adjunct to full-field digital mammography (FFDM) so called combo-mode has been shown to improve sensitivity and reduce false positive rates in breast cancer detection. However, combo-mode screening increases the radiation dose to the patient. In this study, our purpose is to develop a new approach to synthesize photo realistic digital mammogram (SDM) from reconstructed DBT volume to replace adjunct FFDM which in turn will reduce the radiation dose to the patient during breast cancer screening. A deep convolutional neural network (DCNN) is used to synthesize SDM from DBT with FFDM as ground truth during the training. When training our DCNN, a traditional mean squared error (MSE) loss is avoided due to over-smoothing problem. Instead, we propose a gradient guided cGANs (GGGAN) training method to retain subtle tissue structures and microcalcifications (MCs) in the SDM during the DCNN training. The contrast-to-noise ratio (CNR) and the full width at half maximum (FWHM) of the line profiles are used as image quality measures for quantitative comparison. In additional, a human observer perceptual experiment is conducted to qualitatively compare FFDM to SDM of the same patient. The results indicate that SDM has comparable perceptual image quality to FFDM both quantitatively and qualitatively.
KeywordsFFDM DBT Image synthesis Deep learning
- 6.Ledig, C., et al.: Photo-realistic single image super-resolution using a generative adversarial network. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, vol. 2, p. 4 (2017)Google Scholar
- 7.Wang, T.C., et al.: High-resolution image synthesis and semantic manipulation with conditional GANs. arXiv preprint arXiv:1711.11585 (2017)
- 8.Shan, H., et al.: Can deep learning outperform modern commercial CT image reconstruction methods? arXiv preprint arXiv:1811.03691 (2018)
- 10.Chen, L.C., et al.: Rethinking atrous convolution for semantic image segmentation. arXiv preprint arXiv:1706.05587 (2017)
- 11.Ioannou, Y., et al.: Deep roots: improving CNN efficiency with hierarchical filter groups. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 1231–1240 (2017)Google Scholar