Journal of Pediatric Orthopaedics - 2026-04-01 - Journal Article; Validation Study
Development and Validation of an Automated Pipeline for the Detection of Monteggia Fracture Dislocations in Pediatric Radiographs.
Chakladar S, Pereira DE, Tang N, Siddabattula R, Hosseinzadeh P
Topics
Key Takeaway
A two-step deep learning pipeline detected radial head dislocation in pediatric Monteggia fractures with 95–97.5% accuracy and ulnar fracture detection AUC of 0.923, exceeding reported expert interpretation performance.
Summary Depth
Choose how much analysis to show on this article page.
Summary
This study developed and validated a U-Net++ deep learning pipeline to detect Monteggia fracture-dislocations in pediatric elbow radiographs by segmenting the capitellum and radial neck to quantify radiocapitellar misalignment and separately classifying ulnar fractures. Segmentation Dice scores were 0.878 for the capitellum and 0.912 for the radial neck; radial head displacement detection achieved 92.3–100% sensitivity and 96.3–96.9% specificity across lateral and AP views. The ulnar fracture binary classifier reached 87.5% accuracy with AUC 0.923.
Key Limitation
The model was trained and validated on a single-institution dataset of fewer than 500 radiographs without stratification by Bado type, making performance on rarer subtypes (Types II–IV) and across varied imaging systems unknown.
Original Abstract
BACKGROUND
Monteggia fractures are a complex elbow injury that can be missed in up to 50% of pediatric elbow injuries during initial radiographic assessment due to their subtlety. The development of more advanced diagnostic tools is crucial to ensure appropriate treatment for patients. This study investigates whether an automated deep learning pipeline can accurately identify these injuries by quantifying radiocapitellar misalignment and detecting associated ulnar fractures.
METHODS
The automated pipeline utilizes a 2-step, deep learning approach to detect both characteristics of a Monteggia fracture. To identify radial head dislocation, 320 pediatric anteroposterior (AP) and lateral elbow radiographs were used to train a U-Net++ neural network to segment the capitellum and radial neck. Segmentation accuracy was evaluated using the Dice score and intersection over union (IoU). The output segmentations were used to algorithmically construct the radiocapitellar line and measure its displacement from the capitellum. A binary classifier was then trained on 157 paired AP and lateral radiographs to detect the presence of ulnar fractures.
RESULTS
The pipeline displayed high accuracy for the segmentation of both the capitellum (Dice: 0.878 ± 0.098; IoU: 0.794 ± 0.130) and radial neck (Dice: 0.912 ± 0.105; IoU: 0.852 ± 0.142). For the detection of radial head displacement, the model achieved a sensitivity of 92.3%, specificity of 96.3%, and accuracy of 95.0% on lateral elbow radiographs, and a sensitivity of 100.0%, specificity of 96.9%, and accuracy of 97.5% on AP radiographs. The ulnar fracture detection model achieved an accuracy of 87.5% and an area under the curve of 0.923.
CONCLUSION
This study introduces an automated pipeline for the detection of Monteggia fractures in pediatric radiographs. By combining segmentation-based measurement of radial head displacement with binary detection of ulnar fractures, the pipeline demonstrates high diagnostic performance. The model's ability to detect subtle radiographic findings with performance metrics exceeding those of expert interpretation highlights its potential as a clinical decision support tool.
LEVEL OF EVIDENCE
Level III.