5From CT to Custom Implant: The Digital Pipeline

FWhy personalise a prosthesis at all?

Every middle ear is a little different, and a diseased one is more different still. Revision surgery, congenital anomaly, and the bone loss of cholesteatoma all leave behind remnants whose length, angle and position vary widely. Yet the surgeon must bridge the resulting gap with a prosthesis chosen, in theatre, from a tray of a few off-the-shelf lengths and shapes, then trimmed and seated by hand. When reconstruction fails, the cause is often a prosthesis that was the wrong length or shapefor that ear, or that displaced because it never sat securely — not a failure of the surgical idea itself [2023].

The idea behind a custom implant is simple: instead of forcing a generic device to fit, build the device from the patient’s own anatomy. Because middle-ear sound transfer depends on the precise three-dimensional geometry, mass and stiffness of the chain and its coupling to the membrane and stapes, a device that matches the remnant should couple better and stay put [2008]. In a proof-of-concept cadaveric study, four surgeons — blinded to which prosthesis was designed for which ear — correctly matched every printed device to its intended temporal bone; the odds of doing that by chance were about 1 in 1296, evidence that a printed shape really does belong to one specific anatomy [2017]. The promise is better fit, less displacement, and potentially a shorter, less fiddly operation.

FThe pipeline: CT, segment, design, print

Turning a scan into an implant is a four-step digital pipeline, and each step feeds the next:

• Acquire.A high-resolution temporal-bone CT captures the ossicles, footplate and any erosion as a stack of thin slices — the raw data for everything that follows.
• Segment. Those grey-scale voxels are labelled and reconstructed into a three-dimensional surface model (typically an STL file) of the remnant anatomy and the coupling target.
• Design. In computer-aided design (CAD) software the prosthesis is drawn against that surface: shaft length, head shape and coupling footprint matched to the patient.
• Print.The device is built by additive manufacturing — light-cured resin, or printed/sintered titanium, or a bioactive ceramic.

The whole workflow has been demonstrated end-to-end: in the landmark Hirsch study the incus was removed from cadaveric temporal bones, each was scanned with a standard temporal-bone CT protocol, a prosthesis was designed in a commercial segmentation/CAD suite, and the part was printed in photopolymer resin on a desktop stereolithography (SLA) printer [2017]. Step through the stages below.

The single most important thing to understand about this pipeline is that it is data-driven and serial: a design can only be as faithful as the segmentation, and a segmentation can only be as faithful as the scan. Garbage in, garbage out is not a slogan here but a structural property of the workflow.

TImaging and segmentation: the input that limits everything

The pipeline begins with high-resolution computed tomography (HRCT)of the temporal bone — not a routine body CT, whose slice thickness is far too coarse for structures as small as the ossicles. The stapes crura measure well under a millimetre across, so sub-millimetre slices and a high-frequency bone reconstruction kernel are needed to resolve them. CT is chosen over MRI here precisely because it renders fine cortical bone in high contrast; MRI excels at soft tissue (and diffusion-weighted MRI is invaluable for detecting residual cholesteatoma) but does not delineate ossicular bony geometry the way CT does.

How good is that input? Correlating HRCT against intraoperative findings, one prospective series found sensitivity 100%, specificity 88% and accuracy 95% for cholesteatoma, with high accuracy for ossicular erosion overall [2020]. But the correlation was not uniform: it was strong for the malleus and good for the incus, yet weakest for the stapes superstructure and the fallopian canal, where partial erosion and overlying soft tissue blur the picture [2020]. That matters enormously, because the stapes is exactly the structure the prosthesis must couple to.

Segmentation then converts the scan into geometry. An algorithm (or an operator) assigns each voxel to bone, air or soft tissue and reconstructs a surface. Two failure modes dominate: thresholding too aggressively, which thins or breaks fine struts, and mislabelling eroded bone as soft tissue or vice versa. Either error is then baked into the model the implant is designed against. In practice this is why current work emphasises careful, often manually-supervised segmentation and a clear understanding that the printed device inherits every inaccuracy upstream of it [2017, 2025].

TDesigning the implant in CAD

With a faithful surface model in hand, the prosthesis is designed against the patient’s own anatomy. The central design decisions echo conventional ossiculoplasty but are now driven by measured geometry rather than an intraoperative estimate:

• Device type by stapes status. If the stapes superstructure is intact and mobile, the design is a shorter partial ossicular replacement prosthesis (PORP) coupling the membrane or malleus to the stapes head. If the superstructure is absent, the design must reach the footplate as a longer total ossicular replacement prosthesis (TORP) [2025].
• Length and angle. Measured from the segmented model, so the device is neither too short (loose, poor coupling) nor too long (over-tensioned, risking footplate trauma or extrusion).
• Head and footprint. The head plate is shaped to spread contact under the membrane or against the stapes head, and the base to seat stably on the footplate.

A crucial caveat sits inside the design step: a prosthesis is not just a space-filler. Because the middle ear is a tuned mechanical system, the device’s mass, stiffness and alignment shape how efficiently it transmits sound, especially at higher frequencies [2008]. A perfectly anatomy-matched shape that is too heavy or too stiff can still couple poorly. Good custom designs therefore borrow the proven proportions of established titanium PORP/TORP designs — light, slender, broad-headed — and personalise the dimensions within that biomechanical envelope rather than reinventing the shape [2023].

CPrinting: resin, titanium and bioactive ceramics

The final step is additive manufacture, and the choice of process and material is where research is most active. Three routes are emerging:

The most rigorous bench evidence so far comes from vat-photopolymerised PORPs. Modelled on a commercial titanium design and printed in lengths of about 1.5–3.0 mm, they were reproducible at a 0.6 mm shaft, easy to manipulate in cadaveric surgery (if slightly less flexible than titanium), and — critically — gave acoustic transfer similar to a commercial titanium PORP when measured by laser-Doppler vibrometry [2023]. The honest conclusion of that work is measured: it is possible to print functional individualised middle-ear prostheses with good accuracy and reproducibility, and they are currently best suited to surgical training, with clinical use still to be established [2023]. Case-based reports of designing and printing a TORP from the imaging of a cholesteatoma-affected ear point the same direction: a workable workflow, demonstrated in selected defects, not yet a routine clinical product [2025].

CWhere the pipeline helps, and where it cannot

It is worth being clear-eyed about what a custom implant can and cannot do. Its genuine advantages are mechanical and logistical: a device matched to the measured gap should couple more reliably, displace less often, and may shorten the time spent trimming and reseating a generic prosthesis in theatre — the same rationale that motivated the original cadaveric proof-of-concept [2017]. For the difficult ear — revision, marked anatomical distortion, an awkwardly positioned footplate — that is a real benefit.

What the pipeline cannot do is change the biology the device sits in. Ossiculoplasty outcome is dominated not by prosthesis geometry but by middle-ear status: aeration, the health of the mucosa, eustachian-tube function, and the presence of the malleus and a mobile stapes. Prognostic staging systems make this explicit — a poorly aerated, scarred or chronically diseased ear predicts a poor result whatever device is used [2001]. A flawlessly fitted custom implant in a hostile middle ear will still fail, so the pipeline refines fit; it does not replace patient selection.

Two practical points follow for the clinician weighing this technology today:

• The input limits the output.A custom implant is only as good as the CT and the segmentation behind it; the structure hardest to image — the stapes — is the one the prosthesis most depends on, so verify the imaging and the model before trusting the fit [2020].
• The evidence is early and honest. Current data are cadaveric, bench and case-based feasibility, with acoustic parity to titanium but no large clinical trials; printed resin durability and biocompatibility remain open questions, and titanium or bioactive routes may ultimately prove more durable [2023, 2022, 2025].

The digital pipeline is best understood, then, not as a finished product but as a maturing capability: a reproducible way to convert one patient’s imaging into one patient’s implant, already useful for training and proof-of-concept fit, and pointing toward a future in which the middle-ear prosthesis is designed to the ear rather than the ear forced to the prosthesis [2017, 2023].