3D imaging-based AI models outperform demographic models and excel in tibial sizing compared with 2D models in total knee arthroplasty planning: A systematic review.
Authors
Affiliations (4)
Affiliations (4)
- University Department of Orthopedic Surgery and Traumatology, Kantonsspital Baselland, Bruderholz, Switzerland.
- Department of Clinical Research, Research Group Michael T. Hirschmann, Regenerative Medicine & Biomechanics, University of Basel, Basel, Switzerland.
- Karolinska Institute, Stockholm, Sweden.
- Institute of Radiology and Nuclear Medicine, Kantonsspital Baselland, Bruderholz, Switzerland.
Abstract
Accurate preoperative implant sizing is a critical component of successful total knee arthroplasty (TKA). Artificial intelligence (AI) has emerged as a promising tool for enhancing preoperative planning. This is achieved through predictive modelling based on different input modalities, including computed tomography (CT), plain radiographs and patient demographic data. Despite growing interest, the comparative performance of these models remains unclear. This systematic review aims to evaluate and compare the predictive accuracy of AI-based models for TKA component sizing across different input modalities. A systematic literature search was conducted in PubMed, Scopus, Embase and Cochrane Central following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines. Eligible studies included original research that developed or validated AI models for predicting component sizes in TKA planning and reported measurable performance outcomes. Methodological quality and risk of bias were assessed using the PROBAST tool. Qualitative synthesis was performed with stratification by planning modality. Thirteen studies encompassing 37,002 patients met the inclusion criteria. AI models were developed using three-dimensional (3D) imaging (n = 4), two-dimensional (2D) radiographs (n = 4), demographic data (n = 2) or mixed inputs (n = 2). For femoral component prediction, the highest exact size accuracy was achieved by x-ray-based models (86.70%), followed by mixed-input (86.27%), CT/MRI-based (79.98%) and demographic models (45.72%). Accuracy within ±1 size remained high across modalities: CT/MRI (97.67%), x-ray (96.35%) and demographic (92.35%). For tibial components, exact size prediction was similarly high in mixed-input (85.29%), CT/MRI-based (83.98%) and x-ray-based models (83.57%), while demographic models lagged (52.25%). Prediction within ±1 size exceeded 94% for all modalities, with CT-based models achieving the highest accuracy (98.49%). AI models using 2D and 3D imaging achieve high accuracy in TKA component sizing, with 3D imaging performing best for tibial components. Demographic-only models are less accurate, whereas multimodal approaches may optimize predictive precision in surgical planning. Level III.