| Question 1: AI algorithms surveillance post implementation is needed to: |
| Reference: | The ACR Data Science Institute and AI Advisory Group: Harnessing the Power of Artificial Intelligence to Improve Patient Care;McGinty, Geraldine B. et al.; Journal of the American College of Radiology , Volume 15 , Issue 3 , 577 - 579 |
| Choice A: | Ensure the on-going safety and accuracy of the algorithm. |
| Choice B: | Monitor bias. |
| Choice C: | Help vendors improve their algorithms. |
| Choice D: | Provide feedback to facilities about algorithm performance for quality improvement. |
| Choice E: | All of the above. |
| Question 2: The primary goal of validating algorithms is to: |
| Reference: | On fairness and calibration. Geoff Pleiss, Manish Raghavan, Felix Wu, Jon Kleinberg, Kilian Q. Weinberger; 31st Conference on Neural Information Processing Systems (NIPS 2017), Long Beach, CA, USA. |
| Choice A: | Help vendors refine their software. |
| Choice B: | Help train the algorithm. |
| Choice C: | Ensure the algorithm generalizes beyond the training sample and performs without bias. |
| Choice D: | Incorporate clinical knowledge into the software. |
| Question 3: Which of the following is NOT an algorithm used to generate predictive models in machine learning? |
| Reference: | Niedzielski JS, Yang J, Stingo F, Liao Z, Gomez D, Mohan R, Martel M, Briere
T, Court L. A Novel Methodology using CT Imaging Biomarkers to Quantify Radiation
Sensitivity in the Esophagus with Application to Clinical Trials. Sci Rep. 2017
Jul 20;7(1):6034.
|
| Choice A: | Deep Neural Networks. |
| Choice B: | Random Forrest. |
| Choice C: | Support Vector Machines. |
| Choice D: | K-means clustering. |
| Question 4: Which of the following is NOT a concern for Machine Learning algorithms applied to clinical data? |
| Reference: | Valdes G, Solberg TD, Heskel M, Ungar L, Simone CB 2nd. Using machine learning
to predict radiation pneumonitis in patients with stage I non-small cell lung
cancer treated with stereotactic body radiation therapy. Phys Med Biol. 2016 Aug
21;61(16):6105-20.
|
| Choice A: | Availability. |
| Choice B: | Susceptibility to overfitting. |
| Choice C: | Interpretability. |
| Choice D: | Accuracy. |
| Question 5: Which cross validation method is best used to estimate uncertainty ranges of fitted parameters? |
| Reference: | Bradley Efron and Trevor Hastie. Computer Age Statistical Inference. Cambridge: Cambridge University Press. 2017. |
| Choice A: | The Bootstrap. |
| Choice B: | 90/10. |
| Choice C: | The Jack-knife. |
| Question 6: If feature selection is performed prior to a model building step, cross validation should be performed at what point(s)?
|
| Reference: | http://www.alfredo.motta.name/cross-validation-done-wrong/
https://lagunita.stanford.edu/c4x/HumanitiesScience/StatLearning/asset/cv_boot.pdf (pages 17-21)
|
| Choice A: | For set aside data. |
| Choice B: | For iterative results of the modeling building process. |
| Choice C: | Feature selection plus model building. |
| Question 7: Conventional radiomics involves segmentation, feature extraction, dimensionality reduction, and machine learning. “Habitat” imaging obviates the need for which step? |
| Reference: | Gatenby RA, Grove O, Gillies RJ. Quantitative Imaging in Cancer Evolution and Ecology. Radiology 269(1):8-15, 2013. |
| Choice A: | Segmentation. |
| Choice B: | Feature extraction. |
| Choice C: | Dimensionality reduction. |
| Choice D: | Machine learning. |
| Question 8: Conventional radiomics involves segmentation, feature extraction, dimensionality reduction, and machine learning. Deep learning obviates the need for which steps? |
| Reference: | Napel S, Mu W, Perassi B, Aeerts H, Gillies RJ. Quantitative Imaging of Cancer in the Post-genomic Era: Radio(geno)mics, Deep Learning and Habitats. Cancer, 2018 (in press) |
| Choice A: | Segmentation and dimensionality reduction. |
| Choice B: | Segmentation and feature extraction. |
| Choice C: | Segmentation, feature extraction, and dimensionality reduction. |
| Choice D: | Machine learning. |
| Question 9: Which of the following is the main impediment to development of DL-based quantitative imaging tools? |
| Reference: | Steven E. Dilsizian and Eliot L. Siegel. Artificial Intelligence in Medicine and Cardiac Imaging: Harnessing Big Data and Advanced Computing to Provide Personalized Medical Diagnosis and Treatment. Current Cardiology Reports 16:441, 2014.
James A. Brink, Ronald L. Arenson, Thomas M. Grist, Jonathan S. Lewin, Dieter Enzmann. Bits and bytes: the future of radiology lies in informatics and information technology. European Radiology. 27(9):3647–3651, 2017.
|
| Choice A: | Lack of interest in applying quantitative measures in clinical practice. |
| Choice B: | Lack of sufficiently large annotated data sets. |
| Choice C: | Lack of demonstrated value for QI tools. |
| Choice D: | Lack of tools to create good annotations. |
| Choice E: | DL-based tools take too long to apply in clinical practice. |
| Question 10: The only FDA-cleared imaging biomarker is: |
| Reference: | https://www.fda.gov/downloads/Drugs/Guidances/UCM458483.pdf
https://www.fda.gov/drugs/developmentapprovalprocess/drugdevelopmenttoolsqualificationprogram/biomarkerqualificationprogram/ucm535383.htm
|
| Choice A: | Hippocampal volume for Alzheimer disease. |
| Choice B: | Hippocampal volume for temporal lobe epilepsy. |
| Choice C: | Total kidney volume for polycystic kidney disease. |
| Choice D: | Proton density in fatty liver. |
| Choice E: | Relative cerebral blood volume in glioblastoma. |
