Question 1: Which of these is NOT a neural network architecture? |
Reference: | Goodfellow, I., Bengio, Y., & Courville, A. (2016). Deep learning. MIT press. |
Choice A: | Convolutional Neural Network |
Choice B: | Recurrent Neural Network |
Choice C: | Fully Connected Neural Network |
Choice D: | Host Access Neural Network |
Question 2: What are the optimal images for radiation therapy treatment planning? Use CT simulation images as an example: |
Reference: | Barrett, H. H., Kupinski, M. A., Muleller, S., Halpern, H. J., Morris III, J. C., and Dwyer, R., “Objective assessment of image quality vi: Imaging in radiation therapy," Physics in medicine and biology 58(22), 8197 (2013). |
Choice A: | Images acquired with minimal dose. |
Choice B: | Images acquired with as maximum as possible radiation dose allowed by an CT scanner. |
Choice C: | Images acquired with the scanning parameters optimized based on the performance of target contouring methods. |
Choice D: | Images acquired with the scanning parameters optimized based on the treatment outcome. |
Choice E: | Does not matter, CBCT image quality is more important |
Question 3: Which applications can the therapeutic operating characteristic (TOC) curve be employed for? |
Reference: | Barrett, H. H., Myers, K. J., Hoeschen, C., Kupinski, M. A., and Little, M. P., “Task-based measures of image quality and their relation to radiation dose and patient risk," Physics in Medicine & Biology 60(2), R1 (2015). |
Choice A: | Comparison of different imaging instruments |
Choice B: | Comparison of OAR and target delineation, and image registration algorithms |
Choice C: | Comparison of RT treatment planning algorithms |
Choice D: | Determination of optimal treatment dose for individual patients when biological accuracy can be clearly established |
Choice E: | All of the above |
Question 4: Which of these is a potential application of artificial intelligence in treatment planning? |
Reference: | Shen, C., Nguyen, D., Zhou, Z., Jiang, S. B., Dong, B., & Jia, X. (2020). An introduction to deep learning in medical physics: advantages, potential, and challenges. Physics in Medicine & Biology, 65(5), 05TR01. |
Choice A: | Dose/DVH Prediction |
Choice B: | Pareto surface navigation |
Choice C: | DRL for parameter tuning |
Choice D: | Outcome based planning |
Choice E: | All of the above |
Question 5: If the validation of a prediction model shows worse model performance, it means: |
Reference: | van Soest J, Meldolesi E, van Stiphout R, Gatta R, Damiani A, Valentini V, et al. Prospective validation of pathologic complete response models in rectal cancer: Transferability and reproducibility. Med Phys 2017;44:4961–7. https://doi.org/10.1002/mp.12423. |
Choice A: | The model was trained on a biased dataset |
Choice B: | The validation dataset is biased |
Choice C: | We need to know the cohort differences before we can draw any conclusion |
Choice D: | The model needs to be re-learned on the validation dataset |
Question 6: Can toxicity models be re-purposed for different primary tumor locations? |
Reference: | Shi Z, Foley KG, Pablo de Mey J, Spezi E, Whybra P, Crosby T, et al. External Validation of Radiation-Induced Dyspnea Models on Esophageal Cancer Radiotherapy Patients. Front Oncol 2019;9. https://doi.org/10.3389/fonc.2019.01411. |
Choice A: | No, primary tumor locations are leading |
Choice B: | Yes, if the model is properly tested and the AI input is unrelated to the primary tumor |
Choice C: | No, re-purposing AI models is never a good option |
Choice D: | Yes, if the model does work, no adaptation is needed |
Question 7: Which of the following is a valid method for error detection in radiation therapy: |
Reference: | McNutt, T. R., et al. Use of Big Data for Quality Assurance in Radiation Therapy. Seminars in Radiation Oncology 29(4): 326 – 332 (2019). |
Choice A: | Classification Model |
Choice B: | Statistical Outlier |
Choice C: | Random Forest |
Choice D: | Convolutional Neural Networks (CNNs) |
Choice E: | All of the above |
Question 8: The ideal method to evaluate a model for error detection that will be deployed clinically is: |
Reference: | Kalet, A.M., et al. Quality assurance tasks and tools: the many roles of machine learning. Med Phys 47(5): e168-e177 (2019). |
Choice A: | Leave-one-out validation from a single-institution dataset |
Choice B: | Two sets of data from multiple institutions: learning and testing |
Choice C: | Three sets of data from multiple institutions: learning, testing, and validation |
Choice D: | Three sets of data from one institution: learning, testing, and validation |
Question 9: To establish a causal relationship, we need to assume that: |
Reference: | “Targeted Learning: Causal Inference for Observational and Experimental Data.” Mark Van der Laan and Sherri Rose. Springer. 2011. |
Choice A: | There are no hidden confounders and a big enough dataset. |
Choice B: | The covariates are orthonormal to each other. |
Choice C: | The model is interpretable. |
Choice D: | The algorithm is a consistent estimator. |
Question 10: The incorporation of Expert Knowledge into ML algorithms can: |
Reference: | Gennatas, Efstathios D., et al. "Expert-augmented machine learning." Proceedings of the National Academy of Sciences 117.9 (2020): 4571-4577. |
Choice A: | Allow to train with less data. |
Choice B: | Help discover hidden confounders. |
Choice C: | Make models more robust to covariate shifts. |
Choice D: | All of the above. |