3D Printing in Cochlear Implant Research

Robin: Last week we spoke with Iwan and Filip from the University of Cambridge, introducing their groundbreaking work in 3D-printed cochlear models and how their expertise is advancing cochlear implantation techniques. If you missed it, be sure to check out that episode. This week we’re diving deeper: Iwan and Filip return to discuss model accuracy, micro-CT scans, and the challenges of replicating soft tissue. We’ll also explore their work on microfluidic cochlear chips, and the role 3D printing played in shaping their research. Let’s jump in.

Hemdeep: Was there an actual “aha” moment, a point where you felt there was real traction in what you were doing, and the feedback from clinicians suggested you were moving in a positive direction?

Filip: In terms of the research itself, there’s an interesting point here. We were discussing this with the ENT surgeons at Addenbrooke’s Hospital, part of the Cambridge Biomedical Campus, University of Cambridge. It wasn’t really an “aha” moment so much as disbelief that our model could actually follow something like the capstan equation, this fairly simple solution, and have it actually work to describe insertion forces in the cochlea. We spent a lot of time proving that it worked, and there was a lot of back and forth around publishing the paper, partly because there was a belief that other, harder-to-model factors 

 could meaningfully affect the insertion forces too.

The real surprise came when we 3D printed the models, treated them a certain way, and the insertion forces started to look similar to what we saw in cadavers. We knew from the literature and from experience what the insertion forces should look like, and once we found a way to treat the inside of the channel, the lumen, so that the insertion felt and behaved the way it did during cadaveric work, we thought, “we’re onto something.” Then, because we were worried the treatment might change the accuracy of the models, we spent a lot of time developing a technique to double-check the models were still within an acceptable deviation. Once we’d done all that, we felt confident we had one of the best models available to describe the cochlea and do this type of work.

Hemdeep: If the internal model was really the “aha” moment, can you describe the steps you needed to take to develop an accurate internal model? What work came before that?

Filip: First, you really need to start with a good micro-CT scan. We quickly realized a normal CT scan just isn’t good enough for great segmentation. The cochlea is so small that a normal hospital CT scan only gives you about seven slices where the cochlea actually shows up. With micro-CT, it’s about 700, so you can pick up all the features and create a high-resolution model, which is very important for any 3D printing.

Hemdeep: What scale were some of these features at, in terms of millimeters or microns?

Filip: At the entrance, the round window, which is the entry point for the cochlear implant, is about 1 to 1.5 millimeters. As you follow the channel and it starts to spiral, it gets smaller and smaller; by the end, we’re talking about 300 micrometers, maybe 200 to 300 depending on the specific shape. The channel is very small, and the spiral shape is the tricky part.

Iwan: It’s about the size of a pea, we like to say, a very small, tiny structure. There’s a lot of thought that goes into the design too, because once you have the cochlea segmented, it’s sitting in its natural orientation, but to understand the forces in different relevant directions, we have to correct for that within the processing pipeline. We built a whole pipeline in MATLAB, taking the micro-CT scans, segmenting them with template-assisted segmentation, then slicing the cochlea, understanding the cross-sectional area, building an equation for the characteristic spiral, correcting for orientation, and turning it directly into a print, ending with an STL file we can send straight to the printer. That way everything stays consistent.

Depending on how you slice things, for example, we have to remove the little membrane the implant is usually inserted through, since it sits at an angle that would also affect the forces, if we want to isolate just the shape of the cochlea and build up complexity gradually. One key piece of post-processing was improving the transparency of the models, so we could see where the implant was positioned and relate that to the force. Otherwise, in normal anatomy or cadaveric specimens, you’re putting the implant into a small bony channel with no way to see what’s happening inside. The beauty of a 3D-printed model is that you can actually watch what’s happening to the implant: is it bending on itself, is it buckling, and relate that to features you see in the force data, or in the electrical modeling.

Hemdeep: You also mentioned doing some kind of coating, so the internal surface closely mimics real tissue. How did you accomplish that with your printed models?

Filip: We tried a few different coating techniques: soap solutions, medical-grade silicone oil, lots of different things. Then we tested Pluronic, a polymer-based option, and got really good results with that.

Iwan: Yes, a PEG-based surfactant, effectively.

Filip: It helped create a very thin, nice layer, which resulted in a shape and force profile very similar to what we saw in cadavers.

Robin: Did you test other fabrication methods besides 3D printing?

Filip: We were interested in casting for a while; there were papers discussing it. What I quickly disliked is that it takes a lot of time. The process is that you cast material into the cochlear channel within the temporal bone, dissolve the bone itself away, then cast around that resulting part again. It’s more hands-on than 3D printing, and that introduced a lot of variability; creating the same model ten times could produce real deviation, since so much of it is manual work. That was something we worried about with post-processing, specifically the coating, since it was done by hand. But we did a lot of analysis using micro-CT again, rescanning what we’d printed and coated, and comparing it to the original CAD model to check the actual deviation, and we were pleased to find it was minimal for our use case.

Iwan: And the issue with casting is that it’s destructive; once you make that one model, that’s it. With 3D printing, once you have the file, you can manipulate it, add different channels for flushing water or whatever else you need, and replicate the file as many times as you want. Using that nominal-versus-actual analysis Filip mentioned, we got down to about 30 microns of surface deviation, which was the resolution of the printer we were using, the CADWorks printer. We were very happy with that level of accuracy.

We also looked at classical stereolithography techniques, more to build a cochlear-on-a-chip model: a microfluidic version of the cochlea, where we wanted a flat structure we could inject cells into, let them inhabit the model as they would in real anatomy, and then see how they respond electrically when stimulated by a cochlear implant. The issue there was that although it could probably get you smaller, finer features, you couldn’t get the size of a one-millimeter-high channel, which is very difficult to make with that technique. It’s a strange scale to work with: not small enough for a lot of microfabrication techniques, but not big enough for standard 3D printers or other fabrication techniques either. It sits in this awkward middle scale that’s genuinely hard to engineer for.

Filip: We also tested SLA printers, and to a certain degree it worked. Funnily enough, with the CADWorks printer, even though its stated resolution was sometimes lower on paper (comparing, say, 20 to 30 microns), we were still able to achieve better overall deviation from the model. The deviation across 90% of the model’s surface was actually lower with the CADWorks printer than with other SLA printers, and we studied why, and how DLP was achieving better resolution than expected. One part of it is the material itself: we’re using the CADWorks microfluidic resin, and what we like there is that its viscosity is lower, so it flushes out easily from the channels as they’re being printed. What we found is that with higher-viscosity resins, resin gets trapped inside the channel during printing, and as you add layer after layer, the curing process penetrates a bit further into that trapped resin, creating deviations. That was one of the things we discovered, and we’re planning to publish on it.

Hemdeep: Two things really piqued my interest here. One was that you’d attempted modeling on stereolithography. I understand you weren’t able to get the size you wanted; would you have been able to replicate the work you did on the 3D model on that platform? What part of the work would have translated? What were you actually trying to do there?

Iwan: I can explain a bit more about the cochlear chip we were trying to make. Effectively, we needed a PDMS microfluidic chip we could place on a glass substrate, ideally with a microelectrode array (a set of electrodes patterned onto the glass) that connects to an MEA rig, which can record action potentials from individual auditory neurons. Embedding a lot of electrodes within, and culturing cells within, a 3D print directly is quite difficult, and if you can use glass substrates where everything is already well established, it’s much easier, because you need every advantage you can get with these cell types; extracting cells from auditory neurons is very delicate, and you need to handle and culture them under exactly the right conditions.

So we managed to get a slightly simplified version of the cochlea’s structure into a microfluidic chip, embedding small channels that replicate some of the porosity of bone between the channel and the channel containing the cells. By putting a real cochlear implant inside this microfluidic chip, we could see how the cells reacted electrically. It turned out we focused more on patch clamp electrophysiology, effectively probing a single neuron with a small needle and delivering a defined amount of current, or different current patterns. Other members of the lab worked on this too: Gillies did a lot of the work extracting and culturing the cells, and Paul Galsworth worked a lot on the electrophysiology side, coming from a bioengineering background, alongside Sarantos, who did a lot of the purely computational and engineering work, building out different paradigms of electrical pulses. It was a great collaboration, seeing how you could deliver these different pulse patterns in a high-throughput format, to understand how cells respond to different electrical impulses and make that interface as efficient as possible.

It was a really interesting multidisciplinary project, and it interleaved with some of the more accurate cochlear models we had, where we introduced pores into the full cochlear shape to mimic the cochlea’s electrical properties. We designed different pore structures, again within MATLAB, to create voids in the model, and by filling those voids with saline (which is conductive) versus the resin (which isn’t), we could tune the electrical conductivity of different parts of the structure and mimic real patient electrical profiles. As I mentioned, a lot of Chloe Swords’s work was making specific phantoms that matched cadaveric specimens, comparing cadaveric human tissue directly against these 3D-printed equivalents.

Filip: Basically, by tuning the size of the channels, we could change the current spread within the model. One interesting part of the research is seeing how implant stimulation spreads through the spiral shape. The implant has somewhere between 16 and 22 electrode contacts, and each one is stimulated at a certain frequency, which in turn stimulates a certain group of neurons, mapped to a specific frequency. So what we were studying is: if you stimulate electrode number 13, how does the current spread, and does it also affect neighboring neurons that it shouldn’t?

Hemdeep: Where’s that research at right now? Has it been published, or is it still ongoing?

Filip: Chloe finished her PhD this year, or at the end of last year, and this work is being published as we speak.

Hemdeep: Amazing. Given all the incredible work you’ve done over the years with 3D printing, there must have been some gaps, minor or otherwise, where you thought, “if only we had this, the data would be clearer, with fewer outliers.” How would you address that?

Filip: One key drawback, at least in my research, is that the cochlea, the scala tympani specifically, has this very thin layer of soft tissue inside, which changes the insertion ever so slightly. It would be amazing to have a model that could replicate that soft tissue. Right now, we might try replicating it with multi-material jetting technology, but the problem is the channels are so small that it would likely result in significantly higher deviations. So that’s one gap. Also, looking at multi-material properties, being able to print conductive material directly, that would be very interesting for a lot of reasons, especially for research on cochlear implant stimulation and where the current spreads during stimulation.

Iwan: To pick up on that: to study electrical spread, we’ve focused a lot on designing small channels to manually feed wires into the models, then doing post-scans to see exactly where those wires ended up. But if you could print electrical tracks directly within a 3D structure that small, that would be huge, though it’s a significant challenge. It’s completely understandable why that’s not really possible with current technology, even with multi-material jetting now.

Hemdeep: On our side, I always find the hardware side has a pretty defined trajectory: pushing scale to its limit. The materials side is almost a blend of chemistry and a short-order cook at a fast food restaurant, because the demand, especially from biotech and the broader biology user base, for a wide range of very specific materials is immense. Conductive material in general has become one of the most sought-after categories. But what you’re describing isn’t broad conductivity, it’s being able to embed a thin conductive line within a larger non-conductive matrix. It’d be nice to have something like that. I wish I could just hand you that.

Iwan: Yeah, even just having good flexible materials that replicate the small membranes within the cochlea is a very important problem we’ve really struggled with on the fabrication side. A lot of it we’ve been able to mitigate through design considerations: sandwiching different membranes between prints, adding smaller channels to feed wires or conductive material through, or even adding color to mark where a nerve would sit within a clear model. We’ve found ways around a lot of the limitations, mostly through geometric design. Even with the pore structures I mentioned, we’ve made it so that, working with an otherwise non-conductive material, the way we construct the layout lets us still play with electrical properties. Most of the time, you can find a way around a limitation with some creativity.

Robin: When did 3D printing actually become prominent in cochlear research? I’d assume it’s a fairly emerging technology.

Filip: In the literature, people started experimenting with 3D-printed models a couple of years back, testing different technologies, but they generally didn’t check the deviation from the real model, from a cadaver, so it’s hard to know whether that technology actually worked well or not. 3D printing in this space started with the first SLA printers, around 2015 or so. Creating a model that’s fully characterized, so anyone else can print it, with deviation measured through nominal-versus-actual analysis as proof, that’s fairly unique. It’s been done for other use cases, but as far as we know, not for the cochlea specifically.

Robin: What did models look like before 3D printing? Mostly physical? Cadaveric?

Filip: Exactly that: cadavers, cadaveric temporal bones. One issue is the difference between fresh, freshly frozen, and fixed cadaver specimens. There were also cast models, as we discussed earlier, and milled models. One technique, instead of creating a full 3D model, was to create a 2D model: cast acrylic around a 2D spiral shape and pull out the resulting material, giving you a spiral outline, but not a rising, three-dimensional spiral. The core problem with the cochlea is that it’s too big for microfabrication, but too small for normal 3D printing. We also checked out FDM printing, SLA, and two-photon polymerization, a very high-end microfabrication technique, but again the cochlea is too big for that; it’s great for things like nanoscale lenses, but you need something that can be centimeters in overall size while resolving features below about 50 microns, which is a real challenge.

Robin: Out of curiosity, did either of you have 3D printing experience before working on cochlear implants?

Filip: I had about a year with the aerosol jet printer, which was very specific to that project, and during that year I was also testing out an SLA printer, but not an expensive one.

Iwan: I’d collaborated fairly closely with a friend doing a lot of bioprinting at the time, using extrusion-based printing for polymers and gels embedded with cells; I helped out a bit with imaging and playing around with that. We also had a startup idea around then, mainly for competitions and learning, looking at making a 3D-printed nerve conduit to help bridge nerves after traumatic injury and support regrowth. I focused more on the gels and biomaterials side, he focused on the 3D printing, and others looked at neuron outgrowth. It was an interesting project that sparked my interest in 3D printing, and combined well with my interest in neurotechnology when this opportunity came along.

Filip: I’ve always found that 3D printing lets you design something in CAD and then just have it physically in your hands. That was one thing I really liked about my PhD: it was applied research. We’d design something, and the next day, or within a couple of hours, we had it physically and could see what it actually looked like and how it worked. It’s hard to appreciate how things work in the inner ear, and how small everything is, when you’re working in CAD, where everything is scaled up 200 times and looks huge. Once you 3D print it and see it at one-to-one scale, you really appreciate how delicate the structures are, and how amazing the anatomy of the human ear really is.

Iwan: That’s something we’ve found with the anatomical models we make for education: there’s been huge growth in virtual models, VR and AR, to help people understand anatomy, which is really important for surgeons in terms of spatial relationships, size, and shape. But there’s nothing quite like holding something in your hand that you can actually turn around. You really appreciate the scale, because, as Filip says, you can zoom in as much as you want on a virtual model and still kind of miss the true scale. A lot of the time I’ll design something in CAD and think it’s a certain size, and then realize it’s actually much smaller than I pictured.

Hemdeep: You do get tunnel vision; the blinders go on, and you forget the scale until you print it out, and realize it’s either a lot bigger than you thought, or you genuinely can’t see it, or you just broke it.