Enhancing wrist arthroscopy: artificial intelligence applications for bone structure recognition using machine...

Area of Science:

• Orthopedic Surgery
• Medical Imaging
• Artificial Intelligence

Background:

• Wrist arthroscopy is a complex surgical procedure with a steep learning curve.
• Accurate identification of anatomical structures is crucial but challenging during arthroscopy.
• Artificial intelligence (AI) offers untapped potential for enhancing surgical training.

Purpose of the Study:

• To develop and evaluate an AI algorithm for recognizing carpal bone anatomy during wrist arthroscopy.
• To create a robust dataset of labeled wrist arthroscopy images for algorithm training and validation.

Main Methods:

• A Deeplabv3+ classification algorithm with a U-Net architecture was developed.
• A database of 511 labeled wrist arthroscopy images (4,088 augmented) was created from 20 procedures (10 patient, 10 cadaver).
• Algorithm performance was assessed using a Dice loss score, aiming for >80% accuracy.

Main Results:

• The AI algorithm achieved an average Dice loss score of 89% for carpal bone recognition.
• The system demonstrated reliable detection of various carpal bones.
• Performance varied slightly between different carpal bones.

Conclusions:

• AI can reliably detect carpal bones in wrist arthroscopy, aiding surgical visualization.
• Further algorithm refinement could improve accuracy for all carpal bones.
• Real-world application may enhance arthroscopic wrist surgery training and outcomes.