Abstract
OBJECTIVE: This study aims to develop and internally validate a multivariable logistic regression model and a simplified scoring system, based on standardized ultrasonographic features, for the preoperative differentiation of retroperitoneal ganglioneuroma (GN) from schwannoma (SW), and to evaluate their discrimination, calibration, and clinical utility. METHODS: We retrospectively included patients with retroperitoneal GN or SW confirmed by surgical pathology. Standardized ultrasonographic features were extracted, and candidate predictors were selected using least absolute shrinkage and selection operator (LASSO) regression, while retaining potential confounders (age, sex, lesion long diameter). A multivariable model was constructed, and a six-variable simplified score was derived. Discrimination [area under the curve (AUC)], calibration (intercept, slope, Brier score), and decision curve analysis (DCA) were evaluated using stratified fivefold cross-validation and bootstrap resampling (B = 2,000). Two task-oriented thresholds were predefined: R1 [rule-out, sensitivity (Se) ≥ 0.95] and S1 [standard diagnosis, specificity (Sp) ≥ 0.50]. RESULTS: A total of 74 patients were included (GN, 25, 33.8%; SW, 49, 66.2%). After optimism correction, the multivariable model achieved an AUC of 0.930, and the simplified score achieved an AUC of 0.917. Independent predictors included pelvic extraperitoneal location (loc_pelvic = 1), absence of cystic/necrotic change, and lower SD/LD ratio. For R1, the model threshold of 0.149 yielded Se = 0.960, Sp = 0.837, and negative predictive value (NPV) = 0.976; the score threshold of 0.206 yielded Se = 1.000, Sp = 0.592, and NPV = 1.000. For S1, the model threshold of 0.426 yielded Se = 0.920 and Sp = 0.939, and the score threshold of 0.594 yielded Se = 0.760 and Sp = 0.918. CONCLUSION: Both the multivariable model and the simplified score demonstrated excellent performance in differentiating GN from SW, suggesting potential value as rapid, interpretable tools for bedside use and in resource-limited settings. Their clinical utility should be confirmed through external validation and recalibration in multicenter, prospective cohorts and further enhanced through integration with multimodal imaging such as CT, MRI, and contrast-enhanced ultrasound (CEUS).