| Question 1: Deep learning methods are most effective when applied to large training sets, but in the medical domain large data sets are not always available. |
| Reference: | H. Greenspan, B. van Ginneken and R. M. Summers, "Guest Editorial Deep Learning in Medical Imaging: Overview and Future Promise of an Exciting New Technique," in IEEE Transactions on Medical Imaging, vol. 35, no. 5, pp. 1153-1159, May 2016. |
| Choice A: | True. |
| Choice B: | False. |
| Question 2: Transfer learning: |
| Reference: | S. J. Pan and Q. Yang, "A Survey on Transfer Learning," in IEEE Transactions on Knowledge and Data Engineering, vol. 22, no. 10, pp. 1345-1359, Oct. 2010.
doi: 10.1109/TKDE.2009.191
|
| Choice A: | Allows information to flow from input to output layers. |
| Choice B: | Only allows transfer knowledge between similar applications and data domains. |
| Choice C: | Can save significant amount of labeling effort. |
| Choice D: | All of the above |
| Question 3: Medical images are very similar to natural images |
| Reference: | N. Tajbakhsh et al., "Convolutional Neural Networks for Medical Image Analysis: Full Training or Fine Tuning?," in IEEE Transactions on Medical Imaging, vol. 35, no. 5, pp. 1299-1312, May 2016. doi: 10.1109/TMI.2016.2535302 |
| Choice A: | True. |
| Choice B: | False. |
| Question 4: Training CNN using 2.5D image patches is: |
| Reference: | Roth H.R. et al. (2014) A New 2.5D Representation for Lymph Node Detection Using Random Sets of Deep Convolutional Neural Network Observations. In: Golland P., Hata N., Barillot C., Hornegger J., Howe R. (eds) Medical Image Computing and Computer-Assisted Intervention – MICCAI 2014. MICCAI 2014. Lecture Notes in Computer Science, vol 8673. Springer, Cham. |
| Choice A: | Done by using the center most 3 axial slices. |
| Choice B: | Done by cropping a volume of interest with a margin of 2.5 mm. |
| Choice C: | Done by magnifying image patches 2.5 times. |
| Choice D: | Done by using the axial, coronal and sagittal slices as three image channels. |
| Question 5: To avoid overfitting: |
| Reference: | Chintan Parmar, Joseph D. Barry, Ahmed Hosny, John Quackenbush and Hugo JWL Aerts Data Analysis Strategies in Medical Imaging, Clin Cancer Res March 26 2018 DOI: 10.1158/1078-0432.CCR-18-0385. |
| Choice A: | Shallow (or relatively shallow) networks should be preferred if it doesn’t significantly reduce the performance. |
| Choice B: | Dropout or other regularization methods could be used. |
| Choice C: | Data augmentation could be used. |
| Choice D: | All of the above. |
| Question 6: A U-Net is used for: |
| Reference: | Ronneberger O., Fischer P., Brox T. (2015) U-Net: Convolutional Networks for Biomedical Image Segmentation. In: Navab N., Hornegger J., Wells W., Frangi A. (eds) Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015. Lecture Notes in Computer Science, vol 9351. Springer, Cham |
| Choice A: | Image classification. |
| Choice B: | Image segmentation. |
| Choice C: | All of the above. |
| Choice D: | None of the above. |
| Question 7: The main advantage of a U-Net over other deep learning networks is: |
| Reference: | Ronneberger O., Fischer P., Brox T. (2015) U-Net: Convolutional Networks for Biomedical Image Segmentation. In: Navab N., Hornegger J., Wells W., Frangi |
| Choice A: | The U-Net doesn’t need a GPU for fast training. |
| Choice B: | The U-Net needs less GPU memory then other deep learning networks. |
| Choice C: | The U-Net can be trained from very few images. |
| Question 8: What is the Dice Coefficient used for? |
| Reference: | Dice, L.R., 1945. Measures of the amount of ecologic association between species. Ecology, 26(3), pp.297-302 |
| Choice A: | To describe the similarity of two samples. |
| Choice B: | To describe the probability of a class. |
| Choice C: | To normalize weights between layers. |
| Choice D: | To normalize the probabilities of the network result. |
| Question 9: What is data augmentation used for in deep learning? |
| Reference: | Krizhevsky, A., Sutskever, I. and Hinton, G.E., 2012. Imagenet classification with deep convolutional neural networks. In Advances in neural information processing systems (pp. 1097-1105) |
| Choice A: | To reduce the input data to speed up training. |
| Choice B: | To artificially enlarge the image resolution for better results. |
| Choice C: | To compress the input images. |
| Choice D: | To artificially enlarge the dataset to reduce overfitting on image data. |
| Question 10: What are common methods for data augmentation? |
| Reference: | Krizhevsky, A., Sutskever, I. and Hinton, G.E., 2012. Imagenet classification with deep convolutional neural networks. In Advances in neural information processing systems (pp. 1097-1105) |
| Choice A: | Image rotation. |
| Choice B: | Image flipping. |
| Choice C: | Image translation. |
| Choice D: | Altering color/grey-value intensities of images. |
| Choice E: | All of the above. |
