Dissectible Anatomy Lets Users Cut & Tear Virtual Bodies in Real Time

Medical anatomy applications can display detailed scans, isolate structures, and slice through volumetric data, but interacting with those models rarely resembles working in a dissection lab. Purdue University researchers Tim McGraw and Jack Myers developed Dissectible Anatomy: Embodied Exploration for Education to make virtual bodies considerably more interactive.

The technique converts color cryosection imagery, including data from the Visible Human Project, into deformable models that can be cut, torn, retracted, and separated in real time. Rather than relying on conventional connected meshes or computationally expensive biomechanical simulation, the system combines position-based dynamics, signed distance fields, and volumetric splats processed through portable compute shaders. McGraw and Myers presented the project live at SIGGRAPH 2026, where it received the official Real-Time Live! Audience Choice Award.

We spoke with McGraw about designing virtual models without fixed connectivity, balancing responsiveness against physical accuracy, the technology’s potential for games and VFX, and the uniquely stressful experience of performing a six-minute live technology demonstration in front of thousands of SIGGRAPH attendees.

First, can you please tell me a bit about you, your research, and your background as an introduction?

Tim McGraw, Purdue Professor and Researcher: My Bachelor’s degree was in mechanical engineering and early in my career I designed hydraulic valves and construction equipment. I went back to school and got my MS and PhD in computer science at  the University of Florida. My thesis and dissertation were about medical image processing and visualization. But I always wanted to work in entertainment graphics, so I spent time at a few game studios: Rainbow Studios in Phoenix, EA Tiburon in Orlando and briefly at Schell Games in Pittsburgh.

After that I came to Purdue and have been working here for about 12 years. So, I think this particular project relates well to most of my weird disjointed career.

What originally inspired you to explore interactive virtual dissection, and what limitations did you see in existing anatomy education tools that you wanted to address?

Tim McGraw: I had an interest in the Visible Human Project (VHP) ever since I heard about it as a grad student. At the time the data was collected I think the goal was to do things like real-time virtual dissection, but computers were not quite up to the task yet. By the time graphics hardware caught up, visualization and graphics researchers seemed to have forgotten about it. I am always on the lookout for interesting datasets, so after I did SIGGRAPH Real Time Live! In 2024 (Mesh Mortal Kombat), virtual dissection and surgery simulation seemed to be a natural next step. That original technique only worked on a very coarse voxel grid, so evolving that to handle high resolution 3D textures became the next challenge I decided to work on.

The main limitation we are addressing in existing tools for anatomy education is the lack of interactivity. Many of them use VHP data too, but only let you clip the data to a plane or make structures transparent.  I think modern GPUs make a lot more possible, like deforming and cutting as you would do in a dissection lab.

Your system combines cryosection imagery, position-based dynamics, signed distance fields, and volumetric splats. Could you explain how those components work together to turn medical imaging data into a model that can be cut, torn, and deformed interactively?

Tim McGraw: All of those things (particles, SDFs, splats) are processed and updated by compute shaders in our codebase. We don’t depend on any game engine, or GPU- or OS-specific libraries, like CUDA. We can get very good hardware acceleration of our technique with compute shaders and make it portable across many devices.

Cryosection images define where we will create PBD particles. We lay a low-resolution grid over the image and create particles at the grid points that cover the specimen. Then we create a 4x higher resolution grid and create splats at those grid points over the specimen. At runtime we update a signed distance value associated with each splat which tells us how far from the visible surface of the model each splat is.

That lets us use signed distance function operations like intersection and subtraction to fracture the model. As the model gets fractured that signed distance function gets updated, so we only draw the outer shell of splats when rendering. 

Rather than pursuing the engineering-level physical accuracy required for surgical simulation, you prioritize visually plausible behavior and real-time performance. How did you determine the right balance between realism, responsiveness, and educational usefulness?

Tim McGraw: We haven’t determined that yet, actually. I just started with what seemed reasonable to me. The real way to determine educational usefulness is to put educators and students in front of it. Do a knowledge pre-test, post-test and then ideally a follow-up post-test to assess knowledge retention. But to do all of that right you need an actual virtual dissection system geared toward education.

All we have right now is a demo of a technique. So that is my current work: talking to educators and getting their input about what I have right now, what other tools do well, what they don’t do well, and incorporating all of that information into a testbed system.

I think that the degree of physical accuracy will not be detrimental to learning outcomes. There are many approximations used in game physics and collision detection that don’t get noticed by the player, or if they get noticed they don’t really impact your enjoyment of the game. So we hope the same principle applies here. Our goal isn’t to teach biomechanics of tissue.

We want the tool to teach names of structures, their shapes, relative positions, etc., but within an engaging interactive environment. And within an environment where the lack of consequences gives students the freedom to explore the data. If they make a wrong cut they can just undo it, or if they excised the wrong organ they can just start over. Real dissection should always remain a part of medical training, but a system like this could let early learners get familiar with anatomy and terminology and procedures before entering the actual lab.

What were the greatest technical challenges involved in allowing the model’s topology to change dynamically during cutting or tearing, and why were volumetric splats better suited to this problem than a conventional triangle mesh?

Tim McGraw: Splats are convenient because there is no connectivity to track. When we cut we just move splats from one region to another. With triangles or tetrahedra we would need to cut those. But when you cut a tri/tet the result may not be a tri/tet. And those cuts need to propagate to other adjacent triangles, so it becomes very messy, and particularly tricky in the parallel processing environment of compute shaders.

But to get technical, the biggest challenge was figuring out where to break these processes up into chunks that can run in parallel without read/write conflicts. Sometimes we have to dispatch a shader on the grid of regions, then dispatch on the particles and then dispatch on the regions again to clean things up. Trying to minimize the number of those separate dispatches was and still is a challenge.

Beyond anatomy education, where else could these breakable shape constraints be useful? Could the same underlying technology eventually support real-time soft-body destruction, creature dismemberment, or more detailed material interactions in games and VFX?

Tim McGraw: I think so. VFX pre-vis and in-game destruction effects at a small to medium scale seem doable. Dissectible Anatomy can create models by voxelizing meshes. I am a fan of old horror movies and practical effects used in classic movies like John Carpenter’s The Thing and Sam Raimi’s Evil Dead series.

In Mesh Mortal Kombat, I really wanted to recreate the exploding head effect from Scanners. If this technique does get used in those contexts I hope it is with that same over-the-top almost comedic spirit. And not with real human cryosection images. It would be an interesting art task to design the 3D texture representing alien or monster anatomy.

Tell me about the SIGGRAPH Real Time Live! Show and the process for getting into it and what being in the show is like.

Tim McGraw: There is actually quite a lot to Real Time Live (RTL), but it is all worth it. The submission process is similar to submitting to other SIGGRAPH programs. You upload a demo video and a 6-minute breakdown. Those get reviewed by a jury and a month or so later you may get rejected or conditionally accepted. If you are conditionally accepted there is an online rehearsal you show to the RTL chair and some other jury members. If that goes well, you will be officially accepted a few weeks later. Then you have a month and a half or so to get ready for the show. But during that time you will submit a 3-4 page abstract that gets published, and a short promo video for the conference trailer.

If you are in RTL you have to get to the conference early, which is fun because you see it kind of come to life before it gets crowded. You will show up the Saturday before the conference starts to do the tech setup. When you get there you get to see the room, which is huge. It seats thousands of people. I think it was close to 3000 in LA. There’s some cool room and stage lighting, and rows of desks where each team will sit when they present. This first day is just to work out technical details like video, sound, and network connections.

Then on Sunday, which is the first day of the conference, you get up early for a rehearsal. You run through the whole show, in order, with microphones and lights and the whole sound and video crew. Getting to see a sneak preview of everybody else’s demo is a real perk of the experience. Of course, during this time, teams may be editing code or rewriting their scripts. It’s kind of frantic. You are free to attend the conference and do your own thing on Monday, and then on Tuesday evening the show happens. The evening events are my favorite part of SIGGRAPH: the papers fast-forward, animation theater and RTL.

At the end of a long conference day it is good to sit in a cold, dark room and unwind with more entertainment. Unless you are in the show that night.

You have to get to the stage a few hours before the show. We (me and my student Jack Myers) rehearsed as a team a few times while we waited. When the show time arrives the participants sit with the audience in a few reserved rows near the stage. About two presentations prior to yours you will go backstage, get mic’ed up and wait your turn. It’s nerve-wracking! When it’s your turn, you go out and the lights are super bright. So bright that it’s tough to see the audience.

You do your 6-minute demo, and if you are lucky nothing crashes, and you hear some audience murmuring and laughing from the dark in front of you. Knowing your family is back home watching the live stream on Youtube is a source of comfort and stress at the same time. All that work comes down to 6 exhilarating minutes, then it is suddenly over, and you get to watch the rest of the show with a sense of relief.

At the end there are awards presented for Best In Show (voted on by the RTL jury) and Audience Choice (voted on by the audience). I have been lucky enough to win something both years I participated (2024 and 2026). When it’s over, I feel a bit melancholy because I had been looking forward to it for so long and now it is over. It is a totally different vibe from doing academic presentations at conferences.

I encourage readers with a cool game demo, graphics/visualization technique, or even an interactive physical system, like robotics, to consider submitting.

Anything else you'd like to add?

Tim McGraw: If any anatomy or biology educators are interested in this project, please get in touch. Likewise for anyone with cryosection images they can share. 

Tim McGraw, Ph D., Associate Professor at Purdue University