| Question 1: For the purpose of assessing image quality, the process of creating medical images (referred to as the imaging chain) can be conveniently separated into which two components? |
| Reference: | R.F. Wagner, “Low contrast sensitivity of radiologic, CT, nuclear medicine and ultrasound medical imaging systems,” IEEE Trans. Med. Imag., vol. MI-2, pp. 105–121, 1983. |
| Choice A: | Physics and psychophysics |
| Choice B: | Radiation transport and capture |
| Choice C: | Image acquisition and subsequent processing and display |
| Choice D: | System geometry and image reconstruction |
| Question 2: So-called “Channelized” models were introduced to explain which of the following observations? |
| Reference: | H.H. Barrett, J. Yao, J.P. Rolland, and K.J. Myers, “Model Observers for Assessment of Image Quality,” Proc. Nat. Acad. Sci. 90;9758-9765, 1993. |
| Choice A: | Substantially higher human-observer efficiency in low-pass noise compared to high-pass noise. |
| Choice B: | The impact of contrast sensitivity in the human visual system on task performance in noise. |
| Choice C: | Human-observer performance in the Rayleigh discrimination task |
| Choice D: | To quantify the impact of noise sources arising within human observers, which are often referred to as internal noise |
| Question 3: The important property of a “predictive” model observer is to: |
| Reference: | J.G. Brankov, Y. Yang, L. Wei, I. El Naqa, and M.N. Wernick, “Learning a Channelized Observer for Image Quality Assessment,” IEEE Trans. Med. Imag., 28(7), 2009. |
| Choice A: | Fit psychophysical data across a set of imaging conditions. |
| Choice B: | Optimize the performance of imaging systems for the human observer. |
| Choice C: | Help understand the factors that influence human observer performance in simple detection and discrimination tasks in noisy images. |
| Choice D: | Accurately generalize to new imaging conditions for which there may not be any psychophysical data available. |
| Question 4: Which of the following statements are true regarding the relationship between task-based measures of image quality and dose (mark all that apply): |
| Reference: | Barrett, H.H., Myers, K.J., Hoeschen, C., Kupinski, M.A. and Little, M.P., 2015. Task-based measures of image quality and their relation to radiation dose and patient risk. Physics in Medicine & Biology, 60(2), p.R1. |
| Choice A: | Task-based measures of image quality do not vary with patient dose for ionizing imaging system. |
| Choice B: | Decreasing patient dose generally improves image quality |
| Choice C: | Increasing patient dose generally improves image quality. However, there are no agreed upon methods for combining task-based measures of image quality with dose. |
| Question 5: The key difference between a receiver-operating characteristic (ROC) curve and a therapy-operating characteristic (TOC) curve is: |
| Reference: | Barrett, H.H., Kupinski, M.A., Müeller, S., Halpern, H.J., Morris III, J.C. and Dwyer, R., 2013. Objective assessment of image quality VI: Imaging in radiation therapy. Physics in Medicine & Biology, 58(22), p.8197. |
| Choice A: | A TOC curve can be generated for a single patient whereas a ROC curve requires an ensemble of patients |
| Choice B: | The area under the curve can be used as a figure of merit for ROC curves but not TOC curves |
| Choice C: | A TOC curve has probability axes whereas an ROC curve does not |
| Choice D: | None of the above |
| Question 6: Which of the following statements is true regarding targeted alpha therapy |
| Reference: | Kim, Y.S. and Brechbiel, M.W., 2012. An overview of targeted alpha therapy. Tumor biology, 33(3), pp.573-590. |
| Choice A: | Alpha particles travel a very short distance and impart their dose very locally |
| Choice B: | Alpha particles impart > 400x cytotoxicity compared to x-rays or beta rays |
| Choice C: | Alpha emitting isotopes are likely to disassociate from the targeting ligand after emission due to a high recoil |
| Choice D: | The parent and progeny may have corresponding gamma-emission spectra which allows for imaging of the various isotope distributions |
| Choice E: | All of the above |
| Question 7: The application of a deep learning-based image denoising or super-resolution method to medical image data is guaranteed to improve task-based image quality measures. |
| Reference: | (1) Li, Kaiyan, Weimin Zhou, Hua Li, and Mark A. Anastasio. "Task-based performance evaluation of deep neural network-based image denoising." In Medical Imaging 2021: Image Perception, Observer Performance, and Technology Assessment, vol. 11599, p. 115990L. International Society for Optics and Photonics, 2021.
(2) Kelkar, V.A., Zhang, X., Granstedt, J., Li, H. and Anastasio, M.A., 2021, February. Task-based evaluation of deep image super-resolution in medical imaging. In Medical Imaging 2021: Image Perception, Observer Performance, and Technology Assessment (Vol. 11599, p. 115990X). International Society for Optics and Photonics. |
| Choice A: | True |
| Choice B: | False |
| Question 8: Deep learning-based image restoration methods cannot improve the performance of the ideal observer because: |
| Reference: | Foundations of Image Science, Barrett &Myers |
| Choice A: | There may not be enough training data |
| Choice B: | Deep learning models may not have sufficient capacity |
| Choice C: | The ideal observer performance is invariant to non-linear transformations of the image data |
| Question 9: Deep learning-based image restoration methods have the potential to improve traditional IQ measures but degrade signal detection performance because: |
| Reference: | (1) Li, Kaiyan, Weimin Zhou, Hua Li, and Mark A. Anastasio. "Task-based performance evaluation of deep neural network-based image denoising." In Medical Imaging 2021: Image Perception, Observer Performance, and Technology Assessment, vol. 11599, p. 115990L. International Society for Optics and Photonics, 2021.
(2)Kelkar, V.A., Zhang, X., Granstedt, J., Li, H. and Anastasio, M.A., 2021, February. Task-based evaluation of deep image super-resolution in medical imaging. In Medical Imaging 2021: Image Perception, Observer Performance, and Technology Assessment (Vol. 11599, p. 115990X). International Society for Optics and Photonics. |
| Choice A: | The restoration networks can perturb second- and higher-order statistics that are important to observers. |
| Choice B: | Traditional restoration networks are not trained by use of task-informed loss functions. |
| Choice C: | All of the above |
| Question 10: Regarding the impact of the depth of a DNN-based denoising method on the performance of a numerical observer for signal detection, the deep learning mantra “deeper is better” is always true. |
| Reference: | Li, Kaiyan, Weimin Zhou, Hua Li, and Mark A. Anastasio. "Task-based performance evaluation of deep neural network-based image denoising." In Medical Imaging 2021: Image Perception, Observer Performance, and Technology Assessment, vol. 11599, p. 115990L. International Society for Optics and Photonics, 2021. |
| Choice A: | True |
| Choice B: | False |
