Prediction of Retear After Arthroscopic Rotator Cuff Repair Based on Intraoperative Arthroscopic Images Using Deep...

Area of Science:

• Orthopedic surgery
• Medical imaging analysis
• Artificial intelligence in medicine

Background:

• Predicting retear after arthroscopic rotator cuff repair (ARCR) remains a clinical challenge.
• The prognostic value of intraoperative arthroscopic images for ARCR has not been previously analyzed.

Purpose of the Study:

• To evaluate the efficacy of arthroscopic intraoperative images in predicting retear after ARCR using deep learning (DL) algorithms.

Main Methods:

• Retrospective analysis of 1394 arthroscopic images from 580 patients undergoing ARCR.
• Utilized three deep learning architectures (VGG16, DenseNet, Xception) for transfer learning to predict retear.
• Assessed model performance using accuracy, AUC, F1-score, sensitivity, and specificity on training and test sets.

Main Results:

• Deep learning models achieved high accuracy, with DenseNet reaching 91% on the test set.
• DenseNet demonstrated the highest Area Under the Curve (AUC) at 0.92.
• Models showed strong predictive capabilities, with DenseNet achieving 0.93 specificity and 0.84 sensitivity.

Conclusions:

• Deep learning algorithms applied to intraoperative arthroscopic images can accurately predict retear after ARCR.
• This approach offers a promising tool for assessing prognosis and guiding clinical decisions post-surgery.