Artificial Intelligence-Based Solution in Personalized Computer-Aided Arthroscopy of Shoulder Prostheses.

Area of Science:

• Orthopedic Surgery
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate identification of shoulder prostheses is crucial for pre-operative planning, reducing surgical risks and improving patient outcomes.
• Current methods relying on expert interpretation of X-rays are subjective, time-consuming, and error-prone.
• The complexity of shoulder biomechanics necessitates precise implant classification for personalized medicine and apparatus setting.

Purpose of the Study:

• To develop and evaluate deep learning frameworks for automated identification of shoulder implants in X-ray images.
• To introduce the Inception Mobile Fully-Connected Convolutional Network (IMFC-Net) as an efficient deep learning model for this task.
• To compare the performance of IMFC-Net against state-of-the-art models using a public dataset.

Main Methods:

• Three deep learning-based frameworks were proposed, with a focus on the ensemble network IMFC-Net.
• Experiments were conducted on a public dataset of 597 shoulder implants.
• Model generalizability was assessed using augmentation techniques, and interpretability was explored with gradient-weighted class activation mapping.

Main Results:

• IMFC-Net achieved superior performance compared to other models, ranking first in evaluations.
• The model demonstrated high accuracy (89.09%), precision (89.54%), recall (86.57%), and F1-score (87.94%).
• Performance was validated with and without data augmentation, confirming model robustness.

Conclusions:

• The developed IMFC-Net model is highly efficient for classifying shoulder implants from X-ray scans.
• This AI-driven approach can significantly minimize complexities associated with implant revisions.
• The findings support the integration of AI in orthopedic surgery for enhanced decision-making and patient care.