2017 AAPM Annual Meeting
Back to session list

Session Title: Inverse Optimization Meets High Performance Computing
Question 1: FISTA minimizes objective functions of the form f(x) + g(x), where f and g are convex and additionally _____________.
Reference:“Fast gradient-based algorithms for constrained total variation image denoising and deblurring problems,” A. Beck and M. Teboulle, IEEE Transactions on Image Processing, 2009
Choice A:f and g are both differentiable.
Choice B:f has an easy proximal operator and g is unrestricted.
Choice C:f and g both have easy proximal operators.
Choice D:f is differentiable and g has an easy proximal operator.
Question 2: The group sparsity BOO formulation contains dose fidelity terms, a group sparsity term that encourages most candidate beams to be inactive, and an indicator function that enforces nonnegativity constraints. Which of the following are correct:
Reference:“4pi non-coplanar IMRT beam angle selection by convex optimization with group sparsity penalty,” D. O’Connor et al, AAPM 2016 “Beam orientation optimization for intensity modulated radiation therapy using adaptive L2,1-minimization,” X. Jia et al, Physics in Medicine and Biology, 2011
Choice A:The group sparsity term is included in f.
Choice B:The group sparsity term is included in g.
Choice C:The indicator function is included in f.
Choice D:The indicator function is included in g.
Choice E:Both b and d.
Question 3: Volumetric Modulated Arc Therapy (VMAT) creates a single arc by optimizing many closely angularly-spaced beams, and using a version of simulated annealing DAO that:
Reference:Otto, Karl. "Volumetric modulated arc therapy: IMRT in a single gantry arc." Medical physics 35.1 (2008): 310-317.
Choice A:Deterministically solves for the deliverable apertures.
Choice B:Creates multiple deliverable apertures for each beam.
Choice C:Produces fluence maps that then need to be MLC segmented.
Choice D:Solves for exactly one deliverable aperture for each beam.
Choice E:None of the above.
Question 4: The multiphase piecewise-constant Mumford-Shah function:
Reference:Chan, Tony F., Selim Esedoglu, and Mila Nikolova. "Algorithms for finding global minimizers of image segmentation and denoising models." SIAM journal on applied mathematics 66.5 (2006): 1632-1648.
Choice A:Can use a labeling array/function to segment the fluence map.
Choice B:Solves for many overlapping segments.
Choice C:Solves for the segments sequentially, one at a time.
Choice D:All of the above.
Choice E:None of the above.
Question 5: Which statement about the PDHG algorithm is FALSE?
Reference:Chambolle, Antonin, and Thomas Pock. "A first-order primal-dual algorithm for convex problems with applications to imaging." Journal of Mathematical Imaging and Vision 40.1 (2011): 120-145.
Choice A:PDHG updates both a primal variable and a dual variable at each iteration.
Choice B:PDHG is a first-order algorithm.
Choice C:PDHG efficiently solves a system of linear equations at each iteration.
Choice D:PDHG relies on evaluation of the proximal mapping, or “prox operator”.
Choice E:PDHG solves a problem of the form: minimize F(Kx) + G(x).
Question 6: Which of the following is correct?
Reference:J. Kim, W. K. Jeong, and B. Nam, “Exploiting massive parallelism for indexing multi-dimensional datasets on the GPU,” IEEE Transactions on Parallel and Distributed Systems, vol. 26, no. 8, pp. 2258–2271, Aug 2015
Choice A:Irrespective of data size, GPU memory is not a limitation in typical radiotherapy inverse planning implementations.
Choice B:Adding more GPU cards to processing platform will always increase the processing speed.
Choice C:There is an optimal number of GPUs for a computational problem which only depends on the hardware and GPU card specification.
Choice D:There is an optimal number of GPUs for a computational problem which depends on the data size as well as the hardware specifications.
Question 7: Which one is not a typical efficient technique in data management for computational parallelization?
Reference:A. Modiri; X. Gu; A. Hagan; A. Sawant, "Radiotherapy Planning Using an Improved Search Strategy in Particle Swarm Optimization," IEEE Transactions on Biomedical Engineering, vol.PP, no.99, pp.1-1, June 2016.
Choice A:Downsampling.
Choice B:Sparsification.
Choice C:Detecting one-time processes to be performed offline.
Choice D:Increasing the size of data to be shared or transferred between CPU cores or GPUs.
Question 8: Introducing hard dose constraints in an inverse planning process, where a metaheuristic global optimization technique is used, ____________.
Reference:K. Deb, “Multi-Objective Optimization using Evolutionary Algorithms”, WILEY, 2001.
Choice A:Enhances the optimization speed in finding the global solution.
Choice B:Guarantees finding the global solution.
Choice C:Increases the chances of getting trapped in a local solution.
Choice D:Does not change the optimization process, compared to an unconstrained optimization.
Question 9: Why does Monte Carlo dose calculation have a bigger impact in proton therapy compared to photon therapy?
Reference:Paganetti H “Range uncertainties in proton therapy and the role of Monte Carlo simulations.” Physics in Medicine and Biology 2012 57: R99-R117
Choice A:Because proton therapy uses more fields.
Choice B:Because proton therapy uses fewer fields and because of range uncertainties.
Choice C:Because proton therapy because cone-beam CT is not standard in proton therapy.
Choice D:Because proton Monte Carlo codes are more accurate than photon Monte Carlo codes.
Question 10: What is biological optimization in proton therapy?
Reference:Unkelbach J; Botas P; Giantsoudi D; Gorissen B and Paganetti H, “Reoptimization of intensity-modulated proton therapy plans based on linear energy transfer.” International Journal of Radiation Oncology, Biology and Physics 2016 96; 1097-1106
Choice A:Biological optimization tries to reduce dose-averaged LET in the tumor.
Choice B:Biological optimization tries to increase RBE in organs at risk.
Choice C:Biological optimization tries to move elevated RBE values away from critical structures.
Choice D:Biological optimization tries to reduce the number of individual pencils in intensity-modulated proton therapy.
Back to session list