Pigmented villonodular synovitis (PVNS) is a rare but locally aggressive benign proliferative synovial disorder. Its clinical manifestations are highly similar to rheumatoid arthritis (RA), making early radiological differentiation challenging and often leading to misdiagnosis and delayed treatment. To improve early identification of PVNS, this study developed a precise diagnostic model by combining radiomics feature extraction with multiple machine learning algorithms based on magnetic resonance imaging (MRI). We retrospectively enrolled 229 pathologically confirmed knee joint cases (140 PVNS, 89 RA) and acquired images using standardized MRI protocols. A total of 851 high-dimensional imaging features were extracted using PyRadiomics, with 7 most discriminative features retained after selection. Seven classification models were constructed, including logistic regression (LR), random forests (RF), support vector machines (SVM), and XGBoost algorithms. Evaluation metrics evaluation on training and validation sets demonstrated that the XGB model achieved optimal results in the validation set (AUC = 0.80, accuracy = 0.89). Notably, models focusing on synovial tissue as the region of interest showed the highest AUC (average 0.91). The composite model integrating clinical indicators with imaging features further improved diagnostic accuracy to 0.92, outperforming single-modality models. This study confirms that MRI-based machine learning models, particularly the XGB model targeting synovial tissue, can significantly enhance the differential diagnosis efficiency of knee PVNS, providing a feasible clinical decision-support tool.