FP8.8 Deep Learning–Based Automated Detection of Periprosthetic Osteolysis After Primary Total Hip Arthroplasty on Standard Radiographs: A Multicenter Retrospective Study

Clicks: 8
ID: 331298
2026
Free PDF
Article Quality & Performance Metrics
Overall Quality
Not rated
Combines reader engagement with the AI quality analysis. This article has not been analysed, so there is no overall score — reader engagement is measured and shown alongside.
AI Quality Assessment
Not analyzed
Readership in this journal
Emerging

Ranked #417 of 444 articles by views in Journal of hip preservation surgery

Most read Least read

Bar heights use a square-root scale. Only the 120 most-read articles are drawn; the journal has 444 in total.

Mint this article as an NFT
Not yet minted

Create a permanent, verifiable on-chain record of this article on the Scimatic Network. The NFT is held in your Journament account, and you can withdraw it to your own wallet at any time.

5 SUSD one-off · no wallet required
Sign in to mint
Abstract
Abstract ICLA08 - AI and 3D Simulations: Trustworthy Clinical Partners or Tech Hype? Auditorium B (BMCC Level 1), October 8, 2026, 14:00 - 16:00 Background Total hip arthroplasty (THA) is one of the most successful procedures in modern orthopedics. However, aseptic loosening secondary to periprosthetic osteolysis remains a leading cause of long-term implant failure. Radiographic follow-up relies primarily on standard anteroposterior (AP) pelvic radiographs, but early osteolytic changes are often subtle and subject to significant inter-observer variability. Automated image analysis using deep learning may provide a reproducible and objective tool to support radiographic assessment. The aim of this study was to develop and validate a deep learning algorithm for the automatic detection of periprosthetic osteolysis specifically in Gruen zone 1 in patients undergoing primary cementless THA. Methods We conducted a retrospective multicenter observational study including 1,183 primary cementless press-fit THAs performed in 1,039 patients between 2017 and 2024 at two high-volume European centers. Only standard AP pelvic radiographs with a minimum follow-up of 24 months were included. Osteolysis in Gruen zone 1 was independently annotated by experienced orthopedic surgeons, with consensus in cases of disagreement. A two-stage deep learning pipeline was implemented. First, a segmentation network (DeepLabV3 with ResNet-50 backbone) was trained to automatically identify and extract the region corresponding to Gruen zone 1. Second, a binary classification model based on transfer learning (ResNet-18) was trained to detect the presence or absence of osteolysis within the segmented region. To evaluate generalizability, cross-dataset validation was performed by training the model on one institutional cohort and testing it on independent datasets, including a separate cohort from the same center. Performance was assessed using Matthews Correlation Coefficient (MCC) as the primary metric, along with precision, recall, and F1-score. Model interpretability was explored using Grad-CAM activation maps. Results The segmentation network demonstrated stable convergence and good qualitative overlap with manual annotations, enabling reliable extraction of standardized regions of interest. In cross-dataset validation experiments, the classification model achieved MCC values ranging from 0.43 to 0.62. Recall reached up to 0.81, indicating good sensitivity in detecting osteolysis in Gruen zone 1. Performance remained consistent across independent cohorts, suggesting robustness to inter-center variability in imaging acquisition. Grad-CAM visualization confirmed that model predictions were primarily driven by radiographically plausible features located at the bone–implant interface. Conclusions Deep learning–based detection of periprosthetic osteolysis in Gruen zone 1 on standard radiographs is feasible and demonstrates stable cross-dataset performance in a real-world multicenter setting. This approach may serve as a supportive tool for radiographic follow-up of cementless THA, enhancing reproducibility without replacing clinical judgment.
Reference Key
openalex_W7220434306 Use this key to autocite in the manuscript while using SciMatic Manuscript Manager or Thesis Manager
Authors Riccardo Giai Via, Matteo Giachino, Alessandro Massè, Nicolas Bonin, Guido Barducci
Are you one of these authors? Claim your profile
Journal Journal of hip preservation surgery
Year 2026
DOI
10.1093/jhps/hnag043.064
URL
https://doi.org/10.1093/jhps/hnag043.064
Keywords Keywords not found

Citations

No citations found. To add a citation, contact the admin at info@scimatic.org

Comments

No comments yet. Be the first to comment on this article.