2020 Joint AAPM | COMP Virtual Meeting
Back to session list

Session Title: Medical Image Synthesis in Radiotherapy
Question 1: Which of the following is NOT true regarding digital phantoms:
Reference:1. Segars, W. Paul, et al. "4D XCAT phantom for multimodality imaging research." Medical physics 37.9 (2010): 4902-4915. 2. Abadi, Ehsan, et al. "Virtual clinical trial in action: textured XCAT phantoms and scanner-specific CT simulator to characterize noise across CT reconstruction algorithms." Medical Imaging 2018: Physics of Medical Imaging. Vol. 10573. International Society for Optics and Photonics, 2018.
Choice A:Digital phantoms can be used to optimize imaging and treatment techniques in radiology and radiation oncology.
Choice B:Digital phantoms can be used for a virtual clinical trial that is usually carried out after an actual clinical trial.
Choice C:Ideally the digital phantom should simulate all aspects of patient images, including image textures and respiratory motions.
Choice D:The quality of the digital phantoms generated by the deep learning models depends on the quality of the patient data used to train the model.
Question 2: Which of the following is NOT true regarding deep learning-based image quality augmentation methods?
Reference:1. Sahiner, Berkman, et al. "Deep learning in medical imaging and radiation therapy." Medical physics 46.1 (2019): e1-e36. 2. Jiang, Zhuoran, et al. "Augmentation of CBCT Reconstructed From Under-Sampled Projections Using Deep Learning." IEEE transactions on medical imaging 38.11 (2019): 2705-2715.
Choice A:Deep learning models can be used for augmenting various aspects of the image quality, such as reducing the effects of noise, under-sampling and scatter.
Choice B:Training of the deep learning model requires preparation of the training datasets that consist of pairs of low-quality images and the corresponding high-quality ground-truth images
Choice C:Although the training time can be long, prediction time of deep learning models is usually very short, making it applicable for clinical usage.
Choice D:Validation dataset is used to test the final accuracy of the trained deep learning model.
Question 3: Standard distribution matching losses in GAN can cause which one of the following problems:
Reference:Cohen, Luck, Honari. Distribution matching losses can hallucinate features in medical image translation. MICCAI 2018.
Choice A:Removal of structures like tumors
Choice B:Addition of tumors
Choice C:Both a and b
Choice D:Causes blurring
Question 4: GANs are most useful for which of the following problems?
Reference:The reason is because, artifact removal requires corrections learnt over only a small subspace, whereas a task like denoising could be more difficult with even the same modalities exhibiting very different characteristics. With harmonization, although GANs have shown success in producing perceptually good images, they also can introduce artifacts depending on the used loss functions. Armanius, Jiang, Fischer, Kustner, Hepp, Nikolau, Gatidis and Yang. MedGAN: Medical image translation using GANs. Computerized Medical Imaging and Graphics 2020. Wei, Linn, Hsu. Using a generative adversarial network for CT normalization and its impact on radiomics features. ISBI 2020.
Choice A:artifact removal.
Choice B:image harmonization for radiomics analysis.
Choice C:Image denoising.
Question 5: Which of the following methods is likely to yield a more accurate harmonization result:
Reference:This is because a residual connection increase the information available from the encoder layers. Addition of perceptual and style losses will constrain the network to preserve low-level edges, and mid-level textural characteristics. Armanius, Jiang, Fischer, Kustner, Hepp, Nikolau, Gatidis and Yang. MedGAN: Medical image translation using GANs. Computerized Medical Imaging and Graphics 2020. Wei, Linn, Hsu. Using a generative adversarial network for CT normalization and its impact on radiomics features. ISBI 2020.
Choice A:Combat
Choice B:A GAN with residual connections, perceptual and style losses.
Choice C:CNN using mean squared error losses
Choice D:Image resampling
Question 6: Which of the following is correct for MRI-based treatment planning?
Reference:Lei Y, Harms J, Wang T, Liu Y, Shu H, Jani A, Curran W, Mao H, Liu T and Yang X. “MRI-Only Based Synthetic CT Generation Using Dense Cycle Consistent Generative Adversarial Networks,” Medical Physics, 46(8), 3565-3583, 2019.
Choice A:A treatment planning process with MRI as the sole imaging modality will eliminate systematic MR-CT co-registration errors, reduce medical cost, spare the patient from x-ray exposure, and simplify clinical workflow.
Choice B:MRI contains the electron density information, which is necessary for accurate dose calculation.
Choice C:MRI image can be directly used to generate a digitally reconstructed radiograph (DRR) for the current Linac-based treatment setup.
Choice D:All of a, b and c.
Back to session list