Predicting teratoma histology in postchemotherapy residual lesions of non-seminoma testicular cancer (NSTC) patients using integrated CT radiomics and circulating MicroRNAs modelling.

Abstract

5035 Background: Chemotherapy is the primary treatment for metastatic NSTC, but patients often have residual masses afterward. Accurate non-invasive models are needed to predict the histology of these masses, guiding treatment and reserving surgery for those with teratoma. This study aims to enhance predictive accuracy by integrating CT-driven radiomics features with miRNAs 371 and 375 (miR371-375) to distinguish between teratoma and non-teratoma histologies in post-chemotherapy residual masses. Methods: We retrospectively reviewed 52 patients with teratoma (n=56), fibrosis/necrosis (n=34), vGCT (n=11), and seminoma (n=10) lesions, divided into training (N=78) and test (N=33) cohorts with equal class distribution. Lesions included lymph nodes (n=68 retroperitoneum, n=11 mediastinum, n=4 pelvic, n=4 neck), lung (n=21), and brain (n=3) with a median size of 1.6 cm (Q1-Q3 interval=1.2-2.73 cm). Using 3D Slicer version 5.6.1, metastatic masses >1 cm (short axis) were segmented and radiomics features were extracted from venous phase CT images. Plasma miR371 and miR375 levels were measured by RT-PCR before resection. Four machine learning models evaluated the predictive value of radiomics alone (R-only) and combined with miR371/miR375 levels for teratoma histology, and the best performer, Cat Boosting (CB) method, is reported. Results: The analysis of datasets revealed a consistent pattern of superior performance in training sets compared to test sets across all metrics. The CB model R+371+375 dataset demonstrated the most robust overall performance, with the highest AUC values (0.96 [95% CI 0.88-1.0] for training, 0.83 [95% CI 0.68-0.98] for test) and a well-balanced sensitivity (0.71) and specificity (0.76) in the test set for predicting teratoma histology. R+375 followed closely with an AUC of 0.82 (95% CI 0.66-0.97). Conclusions: Combining miR 371 and 375 with CT-driven radiomics features improves the accuracy of classifying teratoma histology in metastatic NSTCs. This method can help characterize teratoma in residual metastatic disease, aiding treatment decisions and minimizing under or over-treatment risks. Further refinement, including the integration of clinical features, will be reported.