| Question 1: Commissioning AI auto-contouring software demonstrates that the software will produce
acceptable contours for clinical use. True or False? |
| Reference: | Commission can be used to demonstrate acceptable performance on a range of data from an
institution. While the commission process can be used to identify combinations of imaging/patient
types for which is unsuitable, the process cannot be relied on to show acceptable performance on
any patient. “All segmentations should be carefully reviewed and approved by the local clini- cal staff
(eg radiation oncologists) before use in a treatment plan.” [Cardenas et al. Advances in autosegmentation.
Seminars in Radiation Oncology 2019;29(3):185-197] |
| Choice A: | true |
| Choice B: | False |
| Question 2: What is the which of these is barrier to clinical adoption of Deep Learning Contouring for OARs? |
| Reference: | Feng M et al. Machine learning in radiation oncology: opportunities, requirements, and needs. Frontiers in oncology. 2018;8:110
e.g Lustberg T et al. Clinical evaluation of atlas and deep learning based automatic contouring for lung cancer. Radiotherapy and Oncology. 2018;126(2):312-7].Cardenas et al. Advances in auto-segmentation. Seminars in Radiation Oncology 2019;29(3):185-197 |
| Choice A: | Not enough evidence/validation of the efficacy of the technology |
| Choice B: | Professional and institutional culture |
| Choice C: | Lack of explainability of AI |
| Choice D: | All of the above |
| Choice E: | None of the above |
| Question 3: Key considerations for the safe deployment of automated treatment planning are |
| Reference: | Training is necessary to educate end users about potential failure modes, the
impact of these failures on patients, and the need for manual review of the plans to prevent these
failures. Manual plan checks, including physician review, is necessary to catch failures. Automated
QA can substantially mitigate the risks of automated planning.
Reference: Kisling et al, A risk assessment of automated treatment planning and recommendations
for clinical deployment, Medical Physics 46(6) 2567-2574, 2019 |
| Choice A: | Training, and automated QA to replace manual plan checks |
| Choice B: | Training, manual plan checks, automated QA are all important |
| Choice C: | Automated planning removes the need for training |
| Choice D: | Automated QA removes the need for training |
| Question 4: Automated contouring based on convolutional neural networks can be used to identify what
percentage of contours that would require minor and major edits is |
| Reference: | The proportion of clinically unacceptable contours that are correctly detected are
99/80%, on average, for contours with minor and major edits.
Reference: Rhee et al, Automatic detection of contouring errors using convolutional neural
networks, Medical Physics 46(11), 5086-5097, 2019 |
| Choice A: | 50% and 90%, respectively |
| Choice B: | 50% and 80%, respectively |
| Choice C: | 80% and 99%, respectively |
| Choice D: | 80% and 90%, respectively |
| Question 5: AI development in medicine/oncology requires |
| Reference: | Patient consent, institutional support, data governance technology, statistical learning algorithms are needed for first data gathering and then for data analysis to develop AI strategies. |
| Choice A: | Patient consent |
| Choice B: | Institutional support |
| Choice C: | Data governance technology |
| Choice D: | Statistical learning algorithms |
| Choice E: | All of the above |
| Question 6: The most important feature in quantitative image analysis for cancer prediction is often |
| Reference: | I. Zhovannik, J. Bussink,A. Dekker ,R. Fijten, R. Monshouwer. Bias in Textural Radiomics. Int J. Rad. Oncol. Biol. Phys VOLUME 105, ISSUE 1, SUPPLEMENT , S118-S119, SEPTEMBER 01, 2019
I. Zhovannik
J. Bussink
A. Dekker
R. Fijten
R. Monshouwer |
| Choice A: | Wavelet textural feature. |
| Choice B: | Intensity histogram. |
| Choice C: | Volume. |
| Choice D: | Sphericity. |
| Choice E: | None of the above. |
