摘要：How to differentiate with MRI-based techniques testicular germ (TGCTs) and testicular non-germ cell tumors (TNGCTs) is still under debate and Radiomics may be the turning key. Our purpose is to investigate the performance of MRI-based Radiomics signatures for the preoperative prediction of testicular neoplasm histology. The aim is twofold: (i), differentiating TGCTs and TNGCTs status and (ii) differentiating seminomas (SGCTs) from non-seminomatous (NSGCTs). Forty-two patients with pathology-proven testicular neoplasms and referred for pre-treatment MRI, were retrospectively enrolled. Thirty-two out of 44 lesions were TGCTs. Twelve out of 44 were TNGCTs or other histologies. Two radiologists segmented the volume of interest on T2-weighted images. Approximately 500 imaging features were extracted. Least Absolute Shrinkage and Selection Operator (LASSO) was applied as method for variable selection. A linear model and a linear support vector machine (SVM) were trained with selected features to assess discrimination scores for the two endpoints. LASSO identified 3 features that were employed to build fivefold validated linear discriminant and linear SVM classifiers for the TGCT-TNGCT endpoint giving an overall accuracy of 89%. Four features were employed to build another SVM for the SGCT-SNGCT endpoint with an overall accuracy of 86%. The data obtained proved that T2-weighted-based Radiomics is a promising tool in the diagnostic workup of testicular neoplasms by discriminating germ cell from non-gem cell tumors, and seminomas from non-seminomas.
其他摘要：Abstract How to differentiate with MRI-based techniques testicular germ (TGCTs) and testicular non-germ cell tumors (TNGCTs) is still under debate and Radiomics may be the turning key. Our purpose is to investigate the performance of MRI-based Radiomics signatures for the preoperative prediction of testicular neoplasm histology. The aim is twofold: (i), differentiating TGCTs and TNGCTs status and (ii) differentiating seminomas (SGCTs) from non-seminomatous (NSGCTs). Forty-two patients with pathology-proven testicular neoplasms and referred for pre-treatment MRI, were retrospectively enrolled. Thirty-two out of 44 lesions were TGCTs. Twelve out of 44 were TNGCTs or other histologies. Two radiologists segmented the volume of interest on T2-weighted images. Approximately 500 imaging features were extracted. Least Absolute Shrinkage and Selection Operator (LASSO) was applied as method for variable selection. A linear model and a linear support vector machine (SVM) were trained with selected features to assess discrimination scores for the two endpoints. LASSO identified 3 features that were employed to build fivefold validated linear discriminant and linear SVM classifiers for the TGCT-TNGCT endpoint giving an overall accuracy of 89%. Four features were employed to build another SVM for the SGCT-SNGCT endpoint with an overall accuracy of 86%. The data obtained proved that T2-weighted-based Radiomics is a promising tool in the diagnostic workup of testicular neoplasms by discriminating germ cell from non-gem cell tumors, and seminomas from non-seminomas.
