| Question 1: Which of the following statements about deep reinforcement learning (DRL) is wrong? |
| Reference: | Human-level control through deep reinforcement learning, V Mnih et al, Nature, 518, pp529–533, 2015 |
| Choice A: | DRL incorporates a deep neural network as agent to interact with the environment |
| Choice B: | DRL automatically generates a number of training samples in a trial-and-error fashion |
| Choice C: | DRL is a type of learning strategies within the scope of deep learning |
| Choice D: | DRL does NOT fall into the regime of deep learning/machine learning |
| Question 2: Why is AI/ML useful for clinical outcome modeling? |
| Reference: | El Naqa I, A Guide to Outcome Modeling in Radiotherapy and Oncology: Listening to the Data, CRC Press Taylor & Francis Group, Boca Raton, FL, 2018. |
| Choice A: | The underlying mechanisms of clinical outcome and patient response are known given current knowledge |
| Choice B: | Machine learning allow for discovering complex patterns in the data such as encountered including radiomics |
| Choice C: | Statistical methods can only be used for testing but not for building such models |
| Choice D: | None of the above |
| Choice E: | All of the above |
| Question 3: Which one of the following measure(s) have been used for clinical evaluation of AI auto-segmentation techniques? |
| Reference: | Caravatta, L., et al., Inter-observer variability of clinical target volume delineation in radiotherapy treatment of pancreatic cancer: a multi-institutional contouring experience. Radiat Oncol, 2014. 9: p. 198. |
| Choice A: | Volume based, e.g., Dice similarity coefficient, Jaccard index |
| Choice B: | Distance based: e.g., Hausdorff distance, mean surface distance |
| Choice C: | Dosimetry based: e.g., DVH changes of the plan |
| Choice D: | All of the above |
| Question 4: Can any known radiobiological model correctly predict local control over the whole range of fractionation schedules? |
| Reference: | Jeong J, Oh JH, Sonke JJ, Belderbos J, Bradley JD, Fontanella AN, Rao SS, Deasy JO. Modeling the cellular response of lung cancer to radiation therapy for a broad range of fractionation schedules. Clinical Cancer Research. 2017 Sep 15;23(18):5469-79. |
| Choice A: | Yes |
| Choice B: | No |
| Question 5: Radiobiological models are always represented as equations that correlate to outcome, True or False? |
| Reference: | Jeong J, Oh JH, Sonke JJ, Belderbos J, Bradley JD, Fontanella AN, Rao SS, Deasy JO. Modeling the cellular response of lung cancer to radiation therapy for a broad range of fractionation schedules. Clinical Cancer Research. 2017 Sep 15;23(18):5469-79. |
| Choice A: | True |
| Choice B: | False |
| Question 6: What are the key challenges facing the development of cancer patient digital twins? |
| Reference: | Hernandez-Boussard T, Macklin P, Greenspan EJ, Gryshuk AL, Stahlberg E, Syeda-Mahmood T, Shmulevich I. Digital twins for predictive oncology will be a paradigm shift for precision cancer care. Nat Med. 2021 Dec;27(12):2065-2066. doi: 10.1038/s41591-021-01558-5. PMID: 34824458. |
| Choice A: | Data challenges such as generating and acquiring high-volume, high-quality, multiscale data |
| Choice B: | Modeling and integration challenges such as seamlessly integrating data-driven and mechanistic modeling |
| Choice C: | Ethical and community challenges such as ethical biases and privacy concerns |
| Choice D: | All of the above |
| Choice E: | None of the above |
