Prediction of discoid lateral meniscus based on tibial plateau anatomical factors and machine learning.

Objectives: The discoid lateral meniscus (DLM) is the most prevalent variant type of the lateral meniscus, and its formation is closely related to specific anatomical features of the tibial plateau. The objective of this study was to systematically evaluate the combined predictive contribution of multiple tibial plateau anatomical parameters and to develop subtype-specific machine-learning models using routinely obtainable radiographic features for the identification of both CDLM and the comparatively understudied ICDLM. Methods: This was a retrospective clinical study, and a total of 494 patients were included, of which the experimental group consisted of 145 patients with complete discoid lateral meniscus (CDLM) and 64 patients with incomplete discoid lateral meniscus (ICDLM), and the control group consisted of 285 patients with normal lateral meniscus (NLM). A number of tibial plateau anatomical parameters were measured radiographically, including tibial plateau width, fibular head height, lateral joint space, height of lateral tibial spine, height of medial tibial spine, lateral slope of the lateral tibial spine, lateral slope of the medial tibial spine, tibial spinous width and tibial eminence width. All data were analyzed by gender grouping in order to compare differences between and within groups. Subsequently, the Least Absolute Shrinkage with Selection Operator (LASSO) regression algorithm was used to screen the feature variables. Seven machine learning prediction models (including: Random Forest, XGBoost, Support Vector Machine (SVM), Logistic Regression, K-Nearest Neighbor (KNN), Naive Bayes, and Gradient Boosting) were constructed based on the screened feature variables, and the best model was selected. Using SHAP (Shapley Additive Explanations) to quantitatively analyze the contribution of each characteristic variable, thereby assessing the relative importance of the characteristic factors and increasing the interpretability of the model. Results: Gender, lateral joint space, height of lateral tibial spine, lateral slope of the lateral tibial spine, lateral slope of the medial tibial spine, and tibial eminence width/tibial plateau width can be used as characteristic parameters of lateral discoid meniscus and lateral incomplete discoid meniscus. Based on these parameters, the best-performing models achieved good discriminative performance for both CDLM and ICDLM in the held-out validation sets, with AUC values of 0.885 and 0.861, respectively. Conclusion: Tibial plateau anatomical parameters may facilitate the identification of DLM. Machine-learning models integrating these parameters provide an efficient radiograph-based approach to support the early recognition of DLM.