Purpose or Case Report: Develop a simple logistic model using only ultrasound report text to predict testicular torsion in 8–18-year-olds when Doppler is inconclusive.
Methods & Materials: We identified 759 pediatric scrotal ultrasound exams from the ED (2019–2024). The modeling cohort included 517 exams ages 8–18, with 68 surgically confirmed torsion cases. From radiology reports we extracted features encoded in text including testes' length, width, and height, hydrocele laterality (if present), scrotal wall edema, and epididymal changes. After preliminary exploratory analysis of the dataset, candidate 3–6 feature logistic models were fit with iteratively reweighted least squares (IRLS). The final model was selected based on AUC, coverage, plausibility, and parsimony. Fivefold stratified cross-validation targeted 95% rule-in specificity and 99% rule-out sensitivity.
Results: The final model contained four features: (1) Aspect ratio (the larger dimension of the short-axis divided by the smaller dimension of the short-axis) of the larger testis; (2) Short axis cross-sectional area of the larger testis; (3) The presence of a hydrocele ipsilateral to the larger testis; (4) Standardized age within the modeling cohort. Overall AUC was 0.893 in 517 patients. With fixed thresholds, the model classified 42.4% of studies, labeling 29.6% as low probability and 12.8% as high probability for testicular torsion. Achieved rule-out sensitivity was 98.5% and rule-in specificity was 94.9%. The aspect ratio was chosen to capture the degree of testis sphericity. In cohort analyses, the aspect ratio of the larger testis was significantly lower in surgically confirmed torsion than in normal examinations (means 1.14 vs 1.33), with strong between-group separation by Welch t-test (p≈1.2e-13) and Mann–Whitney (p≈1.1e-10); effect size was large (Cliff's δ=−0.56), consistent with torsion-related swelling yielding a more spherical short-axis geometry.
Conclusions: A geometry-based model, built from routine report text, accurately stratifies torsion risk in adolescents and may pathophysiologically reflect that torsion-related swelling increases testis sphericity. Under strict operating targets it provides useful rule-in and rule-out bands, supporting decisions when Doppler is limited or equivocal. Prospective, multi-site validation is warranted before clinical deployment.