191 - Submission Reviews

Question	Response
Submission Information:
Id:	papers_0191
Title:	GRIDiron: An interactive authoring and cognitive training foundation for reconstructive plastic surgery procedures
Reviewer #64:
1) Description	This paper presents a physical simulation framework for authoring and training of plastic surgery procedures. Designing a system that works robustly enough to be used in a real-world setting by medical practitioners is very impressive. The surgery tutorials that are included in the video are fascinating. The technical contributions of the paper, however, are not so clear. Of the three contributions listed in line 122, the 1st one is great, but the other two seem a little small, especially the third.
2) Clarity of Exposition	Many parts are good, but I thought that the overall user experience wasn't clear. What were the tasks for the MD residents in the pilot deployment? Did these users mark the incisions? Doesn't CSG take a while (a few minutes)? Were they told to author a surgical procedure? Which parts of their tasks were interactive? Looking at the performance table, the fastest simulation (Rhomboid Flap) takes 0.162 seconds per Newton iteration, which is around 6Hz. How many iterations are required? Or maybe it does not matter too much, since these iterations will appear like damped dynamic steps? These performance numbers are amazing, but I'm just wondering exactly how interactive the sessions were. In the caption for Table 1, it says "Each CG iteration consisted of a single AddDifferential plus a Compact and UnCompact." I expected that summing up these three numbers and multiplying the result by 50 would be the amount of time per Newton iteration, but it doesn't quite add up. Where is the rest of the time being spent? In Figure 1 and in the video at around the 7:09 mark, it seems like the "view" checkbox is selected. Is this session showing only doing a playback of a previously recorded simulation? I think "guided vectorization" needs more than half a page of description in order for us to really appreciate it. Line 261: due to its Line 428: comma not needed Lines 442, 447 (and others): inconsistent capitalization of "section." Line 532: will be determined Lines 599-601: What are external consumers? Figure 7: How are the empty cells determined? It looks like some cells touched by the cut are marked empty, and some are not. Line 702: processing In the code block after line 764, should "Tw" be "scalar_arch" to match 748? Line 772: closing angle bracket missing
3) Quality of References	Fan et al. 2014. Active Volumetric Musculoskeletal Systems. In line 170, it says that FEM was introduced to graphics by O'Brien and Hodgins [1999], but didn't Chen and Zeltzer use it in 1992? Some of the citations could benefit from the use of the shortcite command.
4) Reproducibility	I think it is difficult but possible to reproduce. The limitations are clearly stated in Section 3.
5) Rating	3.5
7) Explanation of Rating	I find the video clips fascinating. I think this is the sort of inter-disciplinary showcase that we all want to achieve. Even though some of the technical contributions seem small, I would be quite happy to see this paper at SIGGRAPH. The scope is clear (Section 3), and there are a lot of interesting insight from the domain experts (e.g., viscoelasticity isn't super important.) In addition to the questions in the "Clarity of Exposition" section, the big question I have is whether a shell simulation would suffice, given that the "computational domain is relatively thin" (line 280). Wouldn't this reduce the DOF quite a bit? Finally, as future work, I think it would be interesting to try out computational plastic surgery: figure out how much to cut or figure out brand new procedures. Maybe the surgeons would object to this.

Reviewer #62:
1) Description	This paper describes a system for simulating plastic surgery procedures for training purposes. The system has been developed with deep involvement from domain experts leading to a working system that makes the right tradeoffs. I am not a simulation expert so it's hard for me to assess the novelty of the system. As an interaction expert I really appreciate the human-centered process of the work and the discussion of tradeoffs throughout the paper. The contributions are both in the geometric representation and in the implementation, both of which have to be handled well to build a working system.
2) Clarity of Exposition	yes.
3) Quality of References	Yes.
4) Reproducibility	The paper does a good job of discussing tradeoffs and explicitly calling out current assumptions. That said, a lot of engineering effort went into building this system. I'm not sure anyone could fit all of the details into a SIGGRAPH paper. This shouldn't count against the paper. It's just a reality of publishing working systems that require a lot of details.
5) Rating	4.0
7) Explanation of Rating	This paper was a pleasure to read. Even though I am not an expert in this area, I found the description fairly accessible and appreciate the context around surgical procedures and training. The paper claims three contributions: a geometric representation, a programming paradigm for optimizing the speed of the simulation computation, and a system architecture. While it's hard for me to evaluate the first two contributions, they seem good advancements to the state of the art. I was less convinced by the claim around the architectural contribution. Perhaps, it's because client-server architectures are so common these days that it makes natural sense to build this type of system with a thin client. The authors are correct, most users (especially students) do not have beefy machines for computation. Offloading the high computation cost to the cloud seems obvious. The separation into two cloud tiers is more interesting - one with specialized hardware and one with lots of memory. But the paper includes little discussion of the various parameters of these two tiers beyond saying that tier 2 needs a lot of memory and should handle all user interaction. I would like to see a deeper discussion here. If a skilled grad student wanted to implement this part, what are the gotchas? I know the paper is already on the long side, but I would have loved to see more discussion of the user feedback from the medical students. There is a bit of it right at the end, but I wanted to know more. How long did they use the system? Was it part of an assignment? Did they learn more effectively thanks to the system? There are always suggestions for improvements, but how close is what you have already. I was very impressed that you used an expert as part of your video. That was really impressive and really made it clear that this work has been developed jointly with surgeons.

Reviewer #47:
1) Description	The manuscript describe a system for simulating surgical procedure of soft tissue manipulation. The system consists of a web-based client, a server for non-simulation process and a numerical simulation server. The manuscript describes the implementation is great details including the engineering aspects and parallelization. The system runs at a near interactive rate for moderate number of voxel.
2) Clarity of Exposition	Yes, but it is too lengthy. Most of section 4 and 5 can be omitted or simply cite other work to shorten the manuscript significantly.
3) Quality of References	Yes.
4) Reproducibility	Yes. Drawbacks and limitations are stated. However, some of them such as self-collisions seem like too critical drawback to have in such a system that goes into great effort to have accurate mechanics. Self-collision problem can be clearly seen in the accompanying video.
5) Rating	2.25
7) Explanation of Rating	I think the end goal of this work could be very beneficial for training medical resident. The 3-tier system is a good idea. However, I feel the manuscript offer little in term of computer graphics technical contributions. The physics simulation technique described in section 4 and 5 are not new. Section 6 is mostly an optimization. The system itself also does not run quite interactively yet. During the first time skimming through the manuscript, I was under the impression that there's rigorous usability being done with medical residents. However, the manuscript seem to only have suggestions that are not yet implemented in the system. The usability study with medical residents would have made the manuscript's contribution stronger. I think that this manuscript is more suitable for a more specialized venue such as MICCAI, ISBMS, MMVR.

Reviewer #66:
1) Description	This paper develops an interactive FEM-based simulation framework targeted as an authoring tool to develop surgical training material. An initial focus here is on a set of reconstructive surgery tasks. See below for an assessment of contributions.
2) Clarity of Exposition	Yes.
3) Quality of References	The references are good. A few important graphics citations in biomedical sim, elasticity and materials that I’m missing are * Li et al, Thin Skin Elastodynamics, TOG 2013 * Sin et al, Vega: Nonlinear FEM Deformable Object Simulator, CGF 2013 * Kharevych et al, Numerical Coarsening of Inhomogeneous Elastic Materials, TOG 2009 * Sueda et al, Musculotendon Simulation for Hand Animation, TOG 2008
4) Reproducibility	As this is a big systems paper I don’t think that this framework could or should be reproduced in its entirety from the paper—-if accepted I’d encourage the authors to provide code. Limitations are pretty well covered.
5) Rating	3.0
7) Explanation of Rating	The goals of this paper are great. Creating an interactive simulation tool that is of practical utility in either surgical planning or (as is the focus here) authoring surgical training material has long been an outstanding research challenge. What’s largely been tricky to date is getting clinicians to actually employ the tools that graphics as a community provides. Thus starting, as this paper does, in the area of using non-predictive simulation to illustrate complex surgical techniques with animation and interaction seems like an exciting and promising research opportunity. That said while the FEM system engineering here is impressive (more on this below) and the client-server framework seems nicely targeted to the application I was pretty disappointed by the lack of user evaluation especially given the evident access to surgical experts and trainees. In particular Sec 7 and 8 seem like a lost opportunity to evaluate and provide evidence of how users in the domain can employ the developed framework and what both success and failure in the user tasks currently look like. In terms of the additional technical contributions the particular simulation model choices seem reasonable and well thought out. These build directly on existing work in graphics (which is made clear in the paper) and the novel variations and extensions appear effective. Given the domain I particularly would have liked to have seen demonstrations of heterogenous material simulation in the training examples. Likewise, a comparison of timing speedup obtained for a consistent set of simulation tasks with the widely available graphics-focused FEM packages like SOFA and VEGA would help build understanding for the speed-up gains obtained by the “final-tier” numerical solver backend developed here.

Reviewer #36:
1) Description	This paper introduces a surgical training system using an embedded elastic simulation method. It presents a practical system that has been deployed to train MD residents. Technically, this paper combines existing embedded simulation techniques, and also proposes a hybrid embedding lattice structure to represent non-manifold tissue topology. Although this paper reuses existing techniques, not proposing significantly new methods, it presents a well-executed example of transferring computer graphics techniques to impact other domains---in this case, surgical training.
2) Clarity of Exposition	The exposition is clear in general. But I found a few details unclear: - When I read section 5 for the first time, I was very confused by the terms, cells and grids. I feel it's better to clarify those terms before presenting the details. - there are many places in the paper where bolded words seem to start a paragraph but get mixed with previous paragraphs. e.g., Line 322, 334, 366, 410, 420, and others. - In sec. 6, it is unclear which part of the system can be run in parallel. It seems the compaction presented in 6.1 is for assembling the stiffness/mass matrices. The nonlinear solve itself cannot be parallelized. - Sec. 6 presents compactation for vectorization using SIMD intrinsics. But it is unclear how the code can be run in multithread over multiple processors. For example, the nodes shared by multiple blocks requires synchronization if naive multithread computation is used. - Guided vectorization in sec. 6.1 uses template classes to handle different types of scalar platforms. But it seems it still requires different and individual template specification for different platforms. Then what is the novelty of this part beyond just some template-based software engineering?
3) Quality of References	Mostly yes. There are many methods in graphics that have used the quasi-static approximation of elastic deformation. I would suggest the authors add: "ARTDEFO: Accurate Real Time Deformable Objects"
4) Reproducibility	This is a large system, so it would be challenging to reproduce it. I appreciate the submission of the code (which I successfully compiled on my computer), which could be helpful for the reproduction.
5) Rating	3.7
7) Explanation of Rating	This is a system paper to integrate elastic simulation techniques into a surgical training system. Technically, the proposed extensions of existing techniques are quite natural but not quite exciting. So I couldn't give a higher score. Nevertheless, it is impressive to see that the authors build a practical system and deployed it for training medical residents. And this paper presents many software engineering ideas to exploit the state-of-the-art hardware and make the system practical. The simplifications introduced in this paper seem reasonable, and the assumptions seem well justified. For that point of view, I think it provides a nice example for graphics community to reach out other domains with what we have developed in our field.