Integrating CT Radiomics and microRNAs to Predict Residual Teratoma in Nonseminomatous Testicular Cancer

Take-Away Points

• ■ Major Focus: To develop a machine learning model integrating CT-derived radiomic features with circulating microRNAs (miR371 and miR375) to predict histologic findings in postchemotherapy residual disease in metastatic nonseminomatous germ cell tumors (mNSGCTs).

• ■ Key Results: A combined model using radiomic features and both microRNAs demonstrated the best performance (area under the receiver operating characteristic curve [AUC]: 0.96 training and 0.83 test sets) in distinguishing teratoma from nonteratoma histologic findings, with balanced sensitivity and specificity.

• ■ Impact: This integrated model may guide decision-making in mNSGCTs, potentially enabling more selective use of surgery by identifying patients likely to have residual teratoma.

Postchemotherapy residual disease in mNSGCTs presents a clinical challenge. While surgery is standard for residual masses larger than 10 mm, nearly half yield benign fibrosis, and the remainder comprise teratoma (approximately 40%) and viable germ cell tumors (vGCTs, approximately 15%). Because biopsies are invasive and prone to sampling error, better noninvasive tools are needed to inform surgical decisions and reduce overtreatment.

Ozgun et al retrospectively developed an integrated model for histologic prediction by combining postchemotherapy CT-based radiomic features with two circulating microRNAs—miR371 (associated with vGCTs) and miR375 (enriched in teratoma)—in patients with residual disease. The study analyzed 111 residual lesions from 52 patients with mNSGCTs, all of whom had posttreatment masses larger than 10 mm and underwent surgical resection with histologic confirmation. Two radiologists independently reviewed CT images and manually segmented residual masses to extract radiomic features.

The authors trained and tested four machine learning models: radiomics alone (R), radiomics plus miR371 (R+371), radiomics plus miR375 (R+375), and radiomics plus both microRNAs (R+371+375). The CatBoost model integrating R+371+375 demonstrated the best, most robust performance for predicting residual teratoma, with AUC values of 0.96 (95% CI: 0.88, 1.0) in the training set and 0.83 (95% CI: 0.68, 0.98) in the test set; sensitivity and specificity were well-balanced.

By effectively distinguishing teratoma from nonteratoma histologic findings, this model could help tailor surgical decisions in patients with mNSGCT with postchemotherapy residual disease. The findings suggest that combining CT-derived radiomic features and circulating miR371 and miR375 improves histologic prediction, offering a step toward personalized management and reduced treatment morbidity.

Highlighted Article

• Ozgun G, Abdalvand N, Ozcan G, et al. Machine learning model integrating computed tomography image–derived radiomics and circulating miRNAs to predict residual teratoma in metastatic nonseminoma testicular cancer. JCO Clin Cancer Inform 2025;9:e2500105. doi: https://doi.org/10.1200/CCI-25-00105