How CT-Based 3D Preoperative Planning Works

Key takeaways

CT based preoperative planning turns a stack of cross-sectional images into a patient-specific surgical plan. The workflow is a pipeline: acquire a CT, read the DICOM data, segment bone into a 3D model, place anatomical landmarks, derive measurements, and build the plan. CT wins over radiographs because it captures size, alignment, and version in three dimensions, the parameters a single 2D projection flattens away. When segmentation runs automatically and the whole thing lives in the browser, the plan that once took hours of manual modelling arrives in minutes, with nothing to install and no images leaving the surgeon's machine.

Why CT, and Why 3D

A radiograph is a single projection: it collapses a three-dimensional bone onto one plane, so magnification, rotation, and overlap all distort what you measure. That is fine for many routine cases, but it hides three things that matter for planning.

• Size in three dimensions. True dimensions, not projected ones, drive implant sizing and resection.
• Alignment. Mechanical-axis angles such as LDFA, MPTA, and HKA are reproducible from a 3D model in a way that varies on a rotated film.
• Version and true offset. Femoral and acetabular version, or humeral version, simply do not exist on a single projection. They live in the axial plane a radiograph cannot show.

This is why CT based 3D planning consistently outperforms 2D templating on sizing accuracy. In one direct-anterior total hip series, exact-size prediction rose from 69% to 88% for the stem and from 56% to 96% for the cup when planning moved from 2D to CT based 3D templating. In robot-assisted total knee arthroplasty, femoral component sizing accuracy went from 52.9% with 2D templating to 96.6% with CT based 3D templating. The added dose of a CT is the trade-off; the payoff is that the plan reflects the real bone. We cover that trade-off directly in 2D templating vs 3D AI planning.

The Pipeline, Step by Step

CT based planning is best understood as a sequence of stages, each feeding the next.

1. Acquire and read the CT (DICOM)

Planning starts with a CT series stored as DICOM, the standard format that carries not just pixels but the geometry: slice spacing, pixel dimensions, and patient orientation. Those tags are what make measurement possible in real-world millimetres rather than screen pixels. If you are new to the format, what is DICOM explains why that metadata is the quiet backbone of every measurement downstream. A good planning tool reads the series, checks orientation, and assembles the slices into a coherent volume.

2. Segment bone into a model

Segmentation is the step that labels which voxels are bone and which are not, one anatomical structure at a time. Done by hand, tracing bone across hundreds of slices is the slow, expensive part, historically hours per case. Modern pipelines automate it with deep learning, and validated models reach Dice overlap above 0.90 against expert contours, with per-patient processing measured in minutes rather than hours. We compare the two approaches in AI vs manual CT bone segmentation, and go deeper on the method in CT bone segmentation for surgical planning.

The output is a clean, patient-specific 3D surface for each bone: femur, tibia, pelvis, or whatever the case requires.

3. Place landmarks

On the 3D model, the system (or the surgeon) marks anatomical landmarks: joint centres, condylar points, mechanical-axis references, tuberosities. These landmarks are the anchors that every downstream measurement is computed from, so their placement defines the plan's accuracy. Automated landmarking speeds this up, but it is exactly the step where surgeon review earns its keep.

4. Derive measurements

From the landmarks, the tool computes the clinically relevant geometry: axis angles, offsets, lengths, and version. The value of doing this on a 3D model is reproducibility. In a validated hip-and-knee CT pipeline, automatically derived measurements showed no statistically significant difference from manual measurements across the tested parameters, with the full per-patient analysis averaging under three minutes. Reproducible numbers are what let a plan be audited and trusted rather than re-argued.

5. Build the plan

Finally, the measurements and model support the surgical plan: sizing, resection levels, alignment targets, positioning. Because the plan is anchored to the patient's own anatomy rather than a population average projected onto a film, it carries into theatre with fewer surprises.

Why Software-Only and Browser-Based Matters

Two architectural choices change what this pipeline feels like to use.

Software-only, implant-agnostic. A plan that does not depend on a specific implant system or a piece of capital hardware travels with the surgeon, not the vendor. The geometry, the model, and the measurements are the product; the implant is data you plug in, not a lock-in.

Browser-based, client-side. When the viewer and the pipeline run in the browser on the surgeon's own machine, there is nothing to install, nothing to maintain, and, in a client-side design, the images can be processed without being uploaded to a third-party server. That has real workflow and data-handling advantages, which we unpack in browser-based DICOM orthopaedic AI. The practical effect is that a heavy 3D workflow stops being a specialist workstation activity and becomes something you open in a tab.

Together these choices are what collapse the cost of the 3D step. When segmentation is automatic and the tool is a web page, the historical barriers to 3D planning, hours of labour and dedicated software, largely disappear.

What a Surgeon Should Expect

CT based 3D planning is a decision-support workflow, not an autopilot. A few realistic expectations:

• You still review. Segmentation and landmarks should be surgeon-checked, especially in deformed, revised, or hardware-laden anatomy where automation is least certain.
• Minutes, not hours. With automated segmentation, expect a per-case turnaround measured in minutes plus your review time, not an afternoon of manual modelling.
• Reproducible numbers. The measurements should be consistent and defensible, which is the whole point of moving off a rotated film.
• Ask the right questions. Before you trust any tool's output, know what to ask about validation, failure modes, and where the human stays in the loop.

FAQ

Do I always need a CT for 3D planning? Yes, 3D planning is CT based. That is why it is best reserved for cases where the added accuracy, sizing certainty, deformity, revision, or version, justifies the dose, rather than applied to every routine primary.

Is the AI making the surgical decision? No. Automated segmentation, landmarking, and measurement are decision support. The surgeon reviews the model and owns the plan; automation removes the manual labour, not the clinical judgement.

How long does the pipeline take? With automated segmentation, per-case processing is typically minutes, plus surgeon review time. The old bottleneck was manual segmentation, which took hours; that is the step automation removes.

Why not just measure on a radiograph? A single projection flattens the third dimension, so version, true offset, and rotation-sensitive angles cannot be measured reliably. CT preserves the real geometry, which is what makes the plan trustworthy. See 2D templating vs 3D AI planning.

The Takeaway

CT based preoperative planning is a pipeline, DICOM to segmentation to model to landmarks to measurements to plan, and each stage is now fast enough to run in minutes in a browser. CT earns its place because it preserves the size, alignment, and version that a radiograph cannot show, and automation plus a software-only, client-side design removes the cost that once made 3D planning a specialist's tool. The surgeon stays in the loop at exactly the steps that matter: reviewing the model, confirming the landmarks, and owning the plan.

Explore the Salnus Surgeon Portal →

---

Disclaimer: This article is for educational and research purposes only. Salnus tools are designated for Research Use Only (RUO) and are not cleared medical devices. Clinical decisions should be made by qualified physicians.

References:

• Accuracy of Preoperative 3D vs 2D Digital Templating for Cementless Total Hip Arthroplasty Using a Direct Anterior Approach. Arthroplasty Today, 2023.
• Preoperative CT-Based Three-Dimensional Templating in Robot-Assisted Total Knee Arthroplasty More Accurately Predicts Implant Sizes than Two-Dimensional Templating. Journal of Knee Surgery, 2019.
• Automating Linear and Angular Measurements for the Hip and Knee After Computed Tomography: Validation of a Three-Stage Deep Learning and Computer Vision-Based Pipeline. Arthroplasty Today, 2024.