Preoperative MRI and clinical indicators for predicting meniscal repairability: a machine learning-based study.

Area of Science:

• Orthopedic Surgery
• Radiology
• Machine Learning

Background:

• Meniscal tears are common knee injuries, and determining repairability preoperatively is crucial for surgical planning.
• Current methods for assessing meniscal repairability often lack precision, leading to suboptimal surgical outcomes.

Purpose of the Study:

• To develop and validate a machine learning (ML) model using preoperative MRI and clinical data to predict the repairability of meniscal tears.
• To identify key imaging and clinical features that predict meniscal repairability.

Main Methods:

• A retrospective analysis of 491 patients who underwent knee MRI and arthroscopy.
• Features extracted from MRI included meniscal morphology, cartilage status, and tear characteristics. Clinical data and demographic variables were also included.
• Multiple ML models (Logistic Regression, Random Forest, GBM, SVM) were trained and evaluated using cross-validation, with performance assessed by AUC, calibration, and DCA.

Main Results:

• The logistic regression model achieved the highest predictive performance (AUC = 0.777).
• Key predictors of non-repairability included ACL injury, high-grade cartilage degeneration, higher BMI, male sex, and greater tear displacement.
• The model demonstrated good calibration and clinical utility across various decision thresholds.

Conclusions:

• A radiology-centered ML model integrating preoperative MRI features can accurately predict meniscal repairability.
• This model shows potential to assist surgeons in optimizing arthroscopic decision-making and surgical planning.
• Further prospective external validation is recommended before widespread clinical adoption.