(Wednesday, 8/1/2018) 10:30 AM - 11:00 AM
Room: Exhibit Hall | Forum 5
Purpose: This paper introduces a deep learning network that restores low-dose CT images of conventionalFBP into CT images of a high quality by semi-supervised learning the image property from iterativereconstruction (IR).
Methods: To guarantee that the resultant high-quality images are consistent with the input images, a CNN-based classi fier is added to the denoising network during the training phase. The classi fier incorporates the CT noise model and evaluates the consistency between the images reconstructed with FBP and those of the denoising network. This supplementary structure makes the entire network class of generative adversarial networks (GANs). For training and testing the network, we use a dataset of 18 patients who have undergone abdominal low-dose CT with both FBP and ASIR, which we split into a training set of 12 patients and a validation set of the remaining 6 patients.
Results: For training and testing the network, we use a dataset of 18 patients who have undergoneabdominal low-dose CT with both FBP and ASIR, which we split into a training set of 12 patients anda validation set of the remaining 6 patients. After being trained with FBP and ASIR image pairs, theGAN successfully recovers the high-quality images from the noisy CT images reconstructed with FBP. It is also remarkable that the GAN successfully preserves the image details, whereas ASIR is known for itsoccasional failure to recover small low-contrast features.
Conclusion: Compared to low-dose CT with FBP, the proposed generative adversarial network signi cantly reduced noise without any loss of small low-contrast features. In addition, our method recovered high-quality images within a second whereas iterative reconstruction methods require much greater computational power.
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email
