CDFAM NYC 2023 · New York · 14–15 June 2023

Scaling & Standardizing Patient Specific Design

Listen as a podcast

Abstract

Presentation Title: Scaling & Standardizing Patient-Specific Design

This recording of the presentation at the CDFAM Computational Design (+DfAM) Symposium, held in New York City in 2023. The speaker is Adam Wentworth, a notable figure in the healthcare and additive manufacturing field, associated with the Mayo Clinic.

Adam Wentworth, a key contributor to the advancements in patient-specific design at the Mayo Clinic, provides an insightful presentation that explores the challenges and solutions associated with scaling and standardizing patient-specific design in healthcare. In his presentation, Wentworth discusses the criticality of personalized healthcare solutions and the role of computational design and additive manufacturing in delivering highly specific and effective treatment options. Drawing from his extensive experience at the Mayo Clinic, Wentworth presents case studies that exemplify the transformative potential of these technologies in healthcare. He explores both the technical and practical aspects of standardizing patient-specific design, offering insights into regulatory considerations, manufacturing processes, and the impact on patient outcomes.

Transcript

From YouTube’s automatic captions, lightly cleaned; expect some errors. Each timestamp opens the video at that moment.

Read the full transcript · 2,459 words

0:05 Thanks everyone, my name is Adam Wentworth, I’m a senior clinical engineer at the Mayo Clinic. I’m going to present our work on automation in patient specific design, and my background is in material science major, with a little bit of 3D printing, and then getting into patient-specific work, and then for the past two and a half years working in a small group at Mayo, and doing very applied 3D printing work in the clinic.

0:37 This is our group, we are housed within the unit of radiology, which is essentially the imaging hub, or the data hub, of a hospital, and so we have access to patient imaging, patient record, and bring together a bunch of CT and MRI data, to really create hundreds of anatomic models and thousands of surgical guides per year, to help treat a variety of pathologies, oncology, and CMF, and all kinds of really interesting work.

1:09 Our group is about 14 or 15 people, split between segmentation engineers, AR VR simulation staff, and radiologists. We’re located in a rural part of southern Minnesota, this is the building we work in, and it houses more CTs than all of Canada, which is a crazy statistic. It has, Mayo Clinic has a very rich history of bringing together engineering and healthcare, and seeing the collaboration between an engineer and a physician really creates quite a bit of magic.

1:46 I’m going to talk about the rationale that we have for doing automation, some of our clinical workflow, that is probably unique to most people’s knowledge space, a little bit on our design, what is holding up, sort of, this performance wise. So we’ve demonstrated this clinical integration of 3D printing for surgical planning, for tumor resections, things like that, in a seamless way, so we’ve basically struggled through scaling this workflow to get to where we are now, which looks small, but you know, it was even smaller before I got there. So, and also proving out the, the effectiveness of this approach, really bringing what they call point of care 3D printing, has significant advantages logistically, and so we’ll kind of talk about a little bit of that.

This bone with a tumor and a guide on it is something that happened on Monday this week, and there’s very rapid turnaround between when an order comes in and when it happens, so that’s another advantage. What you see here is a tumor in a distal tibia, what they’ve done in this specific case is gotten a donor allograft, a donor cadaver bone, digitally we’ve placed that over the affected site and replaced it with a piece of that allograft, and what we’re able to do that digitally is able to have a near perfect match of the cut and the graft position and sizes, and make sure that the performance of a piece like this is just spectacular, that the healing time is really fast, the guides are producing things that speed up operating time, and just make it really easy and really precise.

3:37 This kidney model on the right is one example of some design that we’ve automated, to take CT data of these kidneys, so controlling it from the imaging level and imaging protocol, what they call in radiology, to control all those parameters, to make it optimized for the rest of the workflow, and then outputting a model like this kidney, which is really highly detailed, something that a surgeon would never be able to see from those 2D images, to plan what we call partial nephrectomies, which is sparing a lot of your kidney and preventing people from going on dialysis, so it’s a huge value in someone’s quality of life.

4:15 Unfortunately there’s only a handful of groups that are really 3D printing in the states, we’re at this level, and the, if the technology adoption called the chasm, which is, you know, we’re struggling to figure out, and we have these technology challenges that are enabling most people to just kind of jump into it and get started, so that’s a little bit of what I’ll be talking about. The reasons for that are, one is knowledge, so people don’t have the knowledge of 3D printing and the applications that you’re going to use it for, and a quality management system, and bringing engineers into a hospital, so, and finding engineers to do this kind of work is challenging, because a bioengineer is your typical person that would jump into this space, but they may not have the 3D printing and the design knowledge and everything else that’s kind of required.

4:58 Third is cost, and I think that’s a big thing for a lot of hospitals that are not non-profits like Mayo, is to set up facilities that are like engineering, where you need HVAC, you need a bunch of facilities are not traditional in a hospital, purchasing software, purchasing printers and materials, and justifying it to administration that it’s worth the expense.

5:21 On the second half, we have more regulations, so we have, you know, the FDA was put out a couple papers on, you know, figuring out this space, but obviously they’re a little bit lagging behind the people who are doing it all the time. Reimbursement, right now they do have billing codes to enable some reimbursement for the work that’s done, but they’re temporary, so people are a little bit hesitant to kind of jump into it, because it’s not a guarantee that you’re going to get paid for all of this work. And then you have risk, so we really don’t have the abundance of data that we’ve collected to say where is the risk, how do we prevent it from happening, to the extent that a medical device company has a whole team to kind of throw at this.

6:01 This is our workflow, so segmentation is the part where you take a CT and you apply colors to different parts, it’s really analogous to a lot of other things in machine vision that you would be aware of, so this is starting to be done automatically as well. We have surgical planning, we have design, and this is really where the automation lives, and from segmentation to design, and after that you have the printing and post-processing and stuff that you are probably aware of. Really, the rationale is integration of your DfAM knowledge into the design process, as some companies have already kind of laid out, increasing the value by enabling, upgrading these automated workflows, and also being able to check quality along the way.

6:43 So a knowledge space is pretty much everything you can think of, to design you have to restrict that a little bit in terms of what you’re able to do with software, there are companies that are pushing that limit out, and then you constrain that even more into a control space. This is kind of a pharmaceutical diagram, or concept, but control space is what you can manufacture, then within that control space, so really the DFAM kind of knowledge that goes into it is based on what machines you have, what materials do you have, what performance do you need, from a surgeon’s point of view, and that’s really combining these circles, the designer that probably doesn’t know how to do that surgery, the surgeon, whoever’s using it, and then the technology, and this is our design envelope, this really tiny blue triangle that was already shrunk by all those other constraints.

7:31 So the automation is really to keep it in that small area, and then I’ll talk about human error, so you know, we work on models that have 50 plus parts, where precision is required, quality is required, and then the case might be in a week or so, right, and that’s not typical of a med device company, who can turn around apart in a month or a couple months, they have that time frame to work with. So really, automation is allowing us to overcome those hurdles for a person that doesn’t have two years experience working with the software, or, etc, right.

8:12 We kind of break down our automation into two big categories, some of them are, well, if you’re familiar with coding, and classes of Python, so we can use those codes repeatably for almost every single case. Others are longer workflows that would be specific to an application, it might be a couple hundred steps long, baked in parameters and a couple of options, but really it’s going to do the whole thing.

8:35 The first one is the general use type of script, which is really simple, it’s just coloring stuff, right, it’s a dictionary that has words that it looks up and applies a color. But if you’re doing this manually in the software, it’s like, I don’t know, hundreds of clicks, it’s 10 minutes for a case, right, but doing it in about two seconds can save time, and it can be applied everywhere. And this is what standardization is, it’s me doing it, and another engineer doing it, even a new engineer, and the result is the same, we don’t have to worry about things going wrong.

9:09 Another cool application of coloring is personalization, or customization, so we’re doing this for braces, and having a patient specify what’s their favorite color, type it in, and there’s the output, right, but for an engineer to do this manually would take a very long time, so it’s helping the adoption of new technologies as well. These are actually tiny keychains that our hand therapists can carry around and serve as a little tool to facilitate adoption.

9:35 This is an application specific model, so we have pectus excavatum, so one in one thousand people have this sunken chest, we really want a patient-specific solution for this, the result is very aesthetic in nature. And what we do is, we take these axial images, we draw a normal chest contour on those axial images, and they will produce these molds, to produce these contour guides for the surgeon to bend a titanium bar, fit it into this little half pipe mold, and check the shape before they go put it in a person. These are done routinely, for, for all of these cases now.

10:14 Here’s the kidney type of model, so you start with this as one slice of your axial image, which you can see, they call it worm’s eye view. So the patient’s left is actually on our right, here’s the model that gets produced, this is printed with multi-jet fusion. As part of this workflow, we add these support cones onto it, this enables rapid cooling, right, so we add, we take away like four hours from the print time, and the reason we need these is to support, so the lattice structure of this kidney is not deformed in that process by falling to the bottom, right, or you have this heavy tumor on the inside, it’s going to deform the shape, so that helps us out a little bit. The lattice shape of the kidney also saves about 97 percent of the material versus printing it solid, so you have some material savings too.

11:01 And here’s a applied example, in a unique pathology called a horseshoe kidney, just looks kind of cool, but the automation shows that it works for variable cases, you have variable inputs and you can get these variable outputs. Here’s some parameters capture during, while these scripts are running passively in the background, so we ran, we let it run for the past six months, and then I could go back and grab this data and look at, well, how long does it take on average for a script, after we run 20 of these on different people, real data, right, and what you see is an 80 decrease in time, engineering time, and number of clicks, for these two cases, but we have a lot more. This means an improvement.

11:45 And what’s next, right? So in five years data will be king, data will drive all of these things, so data integration is one of our next ventures, bringing data from an electronic health record, right, where the physicians are used to using this data, they order anatomic models with this system, but the security is crazy high on this, it’s very hard to change, it’s very hard to do anything with that. So we essentially take this data, put it into our manufacturing execution system, right, we’re using co-am right now, we’ve, we’ve developed essentially our dashboard for this, that’s putting out analytics for us, it’s doing a bunch of this data driving, we’re bringing patient data in, we’re gathering print data, we’re getting, gathering effort data.

12:31 It can output a graph like this, which you can’t see, but it is average printing time by technology, it’ll put out a bunch of other stuff too, like average part volumes, so you know, how many small parts, how many big parts. And then lastly, into radiologist dictation, so radiologist has to put all this data together, put in a form, and then they can capture a professional fee for reimbursement, so it’s enabling the value across that chain.

12:57 Nexus AI segmentation, so we’re working with a company in New York, actually, to feed them a bunch of this gold standard, what we call segmentation data, where they can produce a model like this, a spine, that would take a segment or about two hours to create, and they’re producing it in about 10 minutes with superior quality. So this is just the first application it will use AI segmentation for, but it will expand to all of our other applications, I’m sure of it.

13:25 And then AR VR, so you don’t always need a physical model, and if you don’t need a physical model, you can rapidly produce a digital model, send it to the surgeon, even for the patient, console before they go for a surgery, they’re able to see their anatomy, they’re able to communicate with the surgeon on their own, on their own model, ask them really directed questions, and really understand what’s going on with them. This saves all the material costs and all the labor costs, if you don’t, if you don’t need those physical models.

13:55 Just acknowledgments, for, you know, I’m not a programmer, I just had the ideas in the software, we had an intern from Columbia come and do the heavy lifting for us, and did a fantastic job, Jennifer Edinger, and then my predecessors, and people have done some of this automation work before me. So thank you everyone, thank you.

More from CDFAM NYC 2023

Design Computation Human

Design Computation Human

Onur Yüce Gün · New Balance

Mud y Robots: Keynote Presentation

Mud y Robots: Keynote Presentation

Ronald Rael · UC Berkeley / Emerging Objects

Roadmap to Replicators

Roadmap to Replicators

Neil Gershenfeld · MIT Center for Bits and Atoms

Computational Design Ops

Computational Design Ops

Ronnie Parsons · Mode Lab

Overcoming the Design Dilemma

Overcoming the Design Dilemma

Ryan Carter · Johns Hopkins APL

Generative Design & Digital Manufacturing at NASA Goddard

Generative Design & Digital Manufacturing at NASA Goddard

Ryan McClelland · NASA Goddard Space Flight Center

Digital Materials – Geometry and Data Combined

Digital Materials – Geometry and Data Combined

Marek Moffett · General Lattice

Ten Years Teaching Additive

Ten Years Teaching Additive

John Hart · MIT

10 Years of Applying DFAM

10 Years of Applying DFAM

Tim Simpson · Penn State University

MfAM Walks so DfAM Can Run!

MfAM Walks so DfAM Can Run!

Chelsea Cummings · The Barnes Global Advisors

Informatics and AI for Materials Development

Informatics and AI for Materials Development

Wes John Alder · Citrine Informatics

Predicting Fatigue Life in AM Metal Parts Using DigitalClone

Predicting Fatigue Life in AM Metal Parts Using DigitalClone

Behrooz Jalalahmadi · Sentient Science

Designing Material Properties with Advanced Toolpath Generation

Designing Material Properties with Advanced Toolpath Generation

Alex Roschli · Oak Ridge National Laboratory

Design Automation Using Machine Learning & Computational Design

Design Automation Using Machine Learning & Computational Design

Rhushik Matroja · Cognitive Design Systems

Learning to Generate Shapes

Learning to Generate Shapes

Karl Willis · Autodesk Research

Design for Artificial Intelligence

Design for Artificial Intelligence

Chris McComb · Carnegie Mellon University

Register for Updates and Discounts on CDFAM events.