Question 1: The following represent useful applications of deep learning in medical physics and medicine: |
Reference: | Xing L, Krupinski EA, Cai J. Artificial intelligence will soon change the landscape of medical physics research and practice. Med Phys. 2018 Feb 24. doi: 10.1002/mp.12831.
Thrall JH, Li X, Li Q, Cruz C, Do S, Dreyer K, Brink J. Artificial Intelligence and Machine Learning in Radiology: Opportunities, Challenges, Pitfalls, and Criteria for Success. J Am Coll Radiol. 2018 Mar;15(3 Pt B):504-508. doi: 10.1016/j.jacr.2017.12.026. Epub 2018 Feb 4.
|
Choice A: | Image segmentation and classification |
Choice B: | Image analysis and disease diagnosis |
Choice C: | Clinical decision-making |
Choice D: | All of the above |
Question 2: The following are features of deep learning algorithm: |
Reference: | Litjens et al.: A survey on deep learning in medical image analysis, Medical Image Analysis, 2017.
Thrall JH, Li X, Li Q, Cruz C, Do S, Dreyer K, Brink J. Artificial Intelligence and Machine Learning in Radiology: Opportunities, Challenges, Pitfalls, and Criteria for Success. J Am Coll Radiol. 2018 Mar;15(3 Pt B):504-508. doi: 10.1016/j.jacr.2017.12.026. Epub 2018 Feb 4.
|
Choice A: | It uses a large amount of annotated datasets to train a predictive model |
Choice B: | It is generally computationally intensive |
Choice C: | Overfitting may be resulted if the training data is not sufficient |
Choice D: | It is capable of incorporating deep layer of image information into the predictive model |
Choice E: | All of the above |
Question 3: What challenges applicability of deep learning for 3D dose plan analysis: |
Reference: | Toesca et al.: Strategies for prediction and mitigation of radiation-induced liver toxicity, Journal of Radiation Research, 2018
Zhen et al.: Deep convolutional neural network with transfer learning for rectum toxicity prediction in cervical cancer radiotherapy: a feasibility study, Physics in Medicine & Biology, 2017
|
Choice A: | Lack of anatomical information in 3D dose |
Choice B: | Lack of visually distinguishable features in 3D dose plans, i.e. specific corners, edge, etc. |
Choice C: | Different resolution of 3D dose plans for different patients |
Choice D: | Answers A and B |
Choice E: | All of the above |
Question 4: What can be considered the most effective way of simultaneous analysis of 3D dose plans and numeric treatment features (demographics, lab measurements, etc.) by the means of deep learning: |
Reference: | Litjens et al.: A survey on deep learning in medical image analysis, Medical Image Analysis, 2017 |
Choice A: | Converting 3D dose plans and numeric features into 1D feature vector and analyzing the vector with fully-connected neural networks |
Choice B: | Creating a multi-path neural network with convolutional path for 3D dose plans and fully-connected path for numeric treatment features |
Choice C: | Creating individual networks for 3D dose plans and numeric features |
Choice D: | Deep learning does not offer any instruments for analyzing numeric treatment features |
Question 5: Deep learning has been applied to the following breast imaging modalities for classification and/or differentiation of molecular cancer subtypes: |
Reference: | |
Choice A: | Magnetic resonance imaging |
Choice B: | Computed tomography |
Choice C: | Mammography |
Choice D: | Ultrasound imaging |
Choice E: | All of the above |
Question 6: Transfer learning with convolutional neural networks (CNN) has been demonstrated for classification of benign lesions and malignant cancers using magnetic resonance (MR) breast images. The following is correct regarding transfer learning |
Reference: | Kayla R. Mendel, Hui Li, Deepa Sheth, Maryellen L. Giger: Transfer learning with convolutional neural networks for lesion classification on clinical breast tomosynthesis, PROCEEDINGS OF SPIE (SPIE Medical Imaging, 2018, Houston, Texas), 2018.
Benjamin Q. Huynh Hui Li Maryellen L. Giger, Digital mammographic tumor classification using transfer learning from deep convolutional neural networks, Journal of Medical Imaging, 3(3), 034501 (2016).
|
Choice A: | It is often used when there is a lack of sufficient training datasets |
Choice B: | Pre-trained CNN and transfer learning methods enhance the performance in classification of benign and malignant lesions |
Choice C: | It is sometimes done using ImageNet for pre-training of the network |
Choice D: | Answers A and B |
Choice E: | All of the above |