AI in Paediatric Orthopaedics: Why Personalised 3D Planning Is Critical for Growing Bones
Children are not small adults — their growing bones demand surgical precision that generic templates cannot provide. How AI-powered 3D planning, patient-specific instrumentation, and predictive modelling are transforming paediatric deformity correction, DDH screening, and fracture management.
Children Are Not Small Adults
This principle — the foundation of paediatric medicine — carries particular weight in orthopaedic surgery. A child's skeleton is not a scaled-down version of an adult's. It is a dynamic, growing structure governed by growth plates (physes) — the source of future development and the most vulnerable anatomical structure during surgery.
Damaging a growth plate during corrective osteotomy can produce a deformity worse than the one being treated. Angular correction that is acceptable in an adult may lead to overcorrection or undercorrection in a child whose bones will continue growing for years. Implant placement that ignores remaining growth potential may require revision surgery as the child develops.
These challenges make paediatric orthopaedics the subspecialty where personalised, patient-specific surgical planning is not a luxury but a clinical necessity. And it is precisely where AI and 3D planning technologies offer the greatest improvement margin over traditional approaches.
3D Planning in Paediatric Orthopaedics
Why 2D Planning Falls Short
Conventional preoperative planning for paediatric deformity correction relies on 2D radiographs — AP and lateral views that reduce a three-dimensional problem to two dimensions. This approach has known limitations.
Rotational deformities are invisible on standard AP and lateral radiographs. A tibial torsion of 20° may produce a normal-appearing AP view while causing significant functional impairment. Only cross-sectional imaging (CT or MRI) combined with 3D reconstruction reveals the true rotational component.
Multi-planar deformities — simultaneous varus/valgus, flexion/extension, and rotational malalignment — cannot be fully characterised or planned from two orthogonal views. Yet these complex deformities are precisely what paediatric orthopaedic surgeons encounter in conditions like Blount's disease, rickets, and post-traumatic growth arrest.
Growth plate proximity creates a margin of error that 2D planning cannot adequately address. When the osteotomy plane must pass within millimetres of an active physis, the surgeon needs volumetric understanding of the anatomy — not a two-dimensional shadow.
What 3D Planning Enables
A January 2026 study published in Orthopaedic Clinics (Frumberg et al.) documented how 3D planning and patient-specific instrumentation transformed the treatment of a 13-year-old boy with severe tibial deformity. Traditional approaches would have required ankle fusion — an approach that permanently limits mobility. Instead, the surgical team used 3D-printed guides to perform joint-preserving realignment that restored natural ankle mechanics.
The planning process follows a consistent pipeline:
- High-resolution CT acquisition — volumetric imaging of the affected anatomy
- AI-powered bone segmentation — deep learning models extract individual bone surfaces in minutes rather than hours
- Virtual surgical simulation — the surgeon plans the osteotomy on the 3D model, visualising correction angles, hinge points, and growth plate proximity
- PSI design and 3D printing — custom cutting guides translate the virtual plan into intraoperative reality
Where AI Is Making a Difference
Developmental Dysplasia of the Hip (DDH)
DDH screening is one of the most impactful applications of AI in paediatric orthopaedics. Early detection — ideally in the newborn period — prevents the need for complex reconstructive surgery later in childhood.
Deep learning algorithms have been developed to analyse hip ultrasound images, automating the measurement of alpha and beta angles (Graf classification) that determine hip maturity. These models enhance screening accuracy while reducing dependence on the experience of the individual sonographer — particularly valuable in settings where access to specialist paediatric musculoskeletal ultrasonography is limited.
The clinical significance is substantial: DDH that is identified at birth can be treated with a Pavlik harness (non-invasive). DDH diagnosed at walking age may require open surgical reduction and pelvic osteotomy. DDH missed entirely leads to premature osteoarthritis and total hip arthroplasty in young adulthood.
AI does not replace the sonographer — it provides a consistent, quantitative second opinion that catches cases where human interpretation might vary.
Scoliosis: Prediction and Screening
Adolescent idiopathic scoliosis (AIS) management depends critically on predicting curve progression. A curve that will progress to the surgical threshold needs early bracing; a curve that will stabilise needs only observation. Currently, this prediction relies heavily on the Risser sign, chronological age, and the surgeon's experience.
Machine learning models trained on longitudinal scoliosis data have achieved classification accuracy exceeding 90% in predicting which curves will progress. By integrating radiographic parameters (Cobb angle, vertebral rotation), skeletal maturity markers, and patient demographics, these models provide quantitative progression risk that complements clinical judgment.
Smartphone-based scoliosis screening — using deep learning to analyse photographs of the patient's back — represents an emerging approach that could democratise screening — enabling assessment outside specialist clinics.
Bone Age Assessment
Skeletal maturity assessment — traditionally performed by comparing a hand radiograph to the Greulich-Pyle atlas — is fundamental to paediatric orthopaedic decision-making. Bone age determines the timing of growth modulation procedures, epiphysiodesis, and deformity correction surgery.
Deep learning models for automated bone age assessment have been extensively validated and, in several studies, match or outperform the accuracy of experienced paediatric radiologists. These models reduce inter-observer variability and provide instantaneous results — eliminating a bottleneck that can delay surgical planning.
Fracture Detection and Classification
Paediatric fractures present unique challenges for AI. Growth plate injuries (Salter-Harris classification) require specific identification because missed physeal injuries can result in growth arrest and progressive deformity. AI models trained specifically on paediatric radiographs — not adapted from adult fracture detection systems — show promise in identifying these subtle injuries.
Supracondylar humerus fractures, the most common paediatric elbow fracture, have been a particular focus. AI models can detect subtle posterior fat pad signs and anterior humeral line displacement that may be overlooked in a busy emergency department setting.
The Growth Plate Challenge
Why Growth Plates Change Everything
The fundamental difference between paediatric and adult orthopaedic AI is the growth plate. In adults, bone geometry is static — a segmentation model produces a 3D model that accurately represents the anatomy at the time of surgery and will not change.
In children, bone geometry is dynamic. The remaining growth potential — determined by skeletal age (not chronological age) — fundamentally influences the surgical strategy.
Overcorrection in a young child may be intentional (anticipating continued growth in the corrected direction) but catastrophic in an adolescent. Undercorrection may be acceptable if sufficient growth remains to complete the correction naturally, but insufficient if the patient is near maturity.
How AI Can Help
Predictive models that integrate bone age, growth velocity, deformity parameters, and patient demographics can estimate post-surgical growth trajectory. These models do not replace the surgeon's judgment but provide quantitative estimates to inform the correction strategy.
3D planning with AI segmentation enables precise measurement of the distance between the planned osteotomy and the growth plate — a measurement that is difficult to obtain reliably from 2D radiographs alone, particularly in the presence of multi-planar deformity.
From Research to Practice: What Exists Today
Point-of-Care 3D Printing
Hospitals are increasingly establishing in-house 3D printing facilities — "point of care" (PoC) manufacturing — that bring the entire pipeline from CT to printed guide under one roof. A 2023 study documented preliminary results from an in-office 3D printing programme for paediatric limb deformity correction, demonstrating feasibility and cost-effectiveness compared to outsourced manufacturing.
The practical benefit for paediatric cases is significant: same-week turnaround from CT to surgical guide eliminates the 2–4 week delay of external manufacturing, which is particularly important for growing children where timing affects outcome.
AI-Augmented Surgical Planning Platforms
Several preoperative planning platforms now offer paediatric-specific features:
- Growth plate identification and protection zones
- Growth trajectory simulation
- Age-adjusted normal alignment values
- Low-dose CT protocol optimisation
These features represent the convergence of AI segmentation, biomechanical modelling, and clinical paediatric expertise into practical surgical tools.
What Should Paediatric Orthopaedic Surgeons Know?
3D planning is no longer optional for complex deformities. Multi-planar deformity correction in children demands volumetric understanding that 2D planning cannot provide. The investment in CT acquisition and 3D planning pays dividends in surgical precision and reduced reoperation rates.
AI segmentation removes the time barrier. The historical bottleneck — hours of manual segmentation per case — is no longer a valid reason to avoid 3D planning. AI-powered segmentation produces surgical-quality 3D models in minutes.
Growth plate proximity requires special attention. Automated segmentation models trained on adult anatomy may not adequately identify or preserve paediatric growth plates. Verify that any AI tool used for paediatric planning has been validated on paediatric imaging data.
Radiation dose matters. CT acquisition in children carries a higher relative radiation risk than in adults. Low-dose CT protocols optimised for bone segmentation can reduce exposure while maintaining sufficient image quality for AI processing.
Engage early with emerging tools. DDH screening AI, bone age automation, and scoliosis prediction models are maturing rapidly. Early engagement — even with Research Use Only tools for evaluation — positions your practice to adopt validated tools faster when they reach clinical readiness.
Salnus's Approach
While Salnus's current platform focuses on adult knee orthopaedics, our underlying technology pipeline — AI segmentation, STL mesh generation, 3D visualisation — is architecture-ready for paediatric applications. Extension to paediatric-specific models requires dedicated paediatric CT training data with growth plate annotations, which we are exploring through clinical partnerships.
For paediatric orthopaedic surgeons interested in collaborative research on AI-assisted 3D planning, contact our team.
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 regarding paediatric patients should be made by qualified paediatric orthopaedic surgeons based on comprehensive clinical assessment.
References:
- Frumberg D, et al. 3D Planning and Patient-Specific Instrumentation in Pediatric Orthopaedics. Orthop Clin. 2026.
- Vescio A, et al. Artificial Intelligence in Pediatric Orthopedics: A Comprehensive Review. Medicina. 2025;61(6):954.
- Al-Rumaih M, et al. Current Trends in Pediatric Orthopedic Disorders. Cureus. 2026.
- Applications of Artificial Intelligence in Pediatric Orthopaedics. Indian J Orthop. 2025.
- Salnus Research Group. 3D-Printed Patient-Specific Guides for Knee Reconstruction. OJSM. 2026.
Reviewed by the Salnus biomedical engineering team.