Related Experiment Video
Updated: Dec 21, 2025

07:46
Author Spotlight: Revolutionizing Remote Surgery with Augmented Reality and Robotics for Enhanced Precision and Accessibility
Published on: August 9, 2024
1.1K
Automated detection and classification of shoulder arthroplasty models using deep learning
Paul H Yi1,2, Tae Kyung Kim1,2, Jinchi Wei2
1The Russell H. Morgan Department of Radiology and Radiological Science, Johns Hopkins University School of Medicine, Baltimore, MD, USA.
Skeletal Radiology
|May 17, 2020
Summary
Deep convolutional neural networks (DCNNs) accurately detect total shoulder arthroplasty (TSA) implants and differentiate between TSA and reverse TSA (RTSA). These DCNNs can also classify five specific TSA models with high precision, aiding in implant identification.
Area of Science:
- Artificial Intelligence in Medical Imaging
- Orthopedic Surgery
- Machine Learning for Medical Diagnosis
Background:
- Accurate identification of total shoulder arthroplasty (TSA) models is crucial for patient care and research.
- Current methods for identifying TSA models can be time-consuming and require specialized expertise.
- Deep convolutional neural networks (DCNNs) offer a potential solution for automated implant recognition.
Purpose of the Study:
- To develop and evaluate DCNNs for detecting shoulder arthroplasty implants.
- To assess the DCNNs' ability to differentiate between TSA and reverse TSA (RTSA).
- To determine the DCNNs' performance in classifying five specific TSA models.
Main Methods:
- Utilized 482 radiography studies including native shoulders, RTSA, and five TSA models.
- Trained ResNet DCNN binary classifiers for implant detection, TSA vs. RTSA differentiation, and specific TSA model classification.
- Employed 20-fold augmentation, AUC-ROC analysis for performance assessment, and class activation mapping for feature interpretability.
Main Results:
- DCNN achieved an AUC-ROC of 1.0 for implant detection and 0.97 for TSA vs. RTSA differentiation.
- Class activation mapping highlighted the DCNN's focus on characteristic implant components.
- DCNNs classified five TSA models with AUC-ROCs ranging from 0.86 to 1.0, emphasizing unique design features.
Conclusions:
- DCNNs demonstrate high accuracy in identifying TSA implants and distinguishing between TSA and RTSA.
- The developed DCNNs can effectively classify five specific TSA models.
- This work provides a proof of concept for an automated arthroplasty atlas, facilitating rapid model identification.