Table 1.
The optimal machine learning configuration based on the cross-validated AUCs of the training dataset.
Table 2.
Preoperative characteristics of all included patients per dataset.
Table 3.
Final histopathological results of all included patients who underwent radical prostatectomy and extended pelvic lymph node dissection per dataset.
Fig 1.
ROC-curves of the machine learning models using radiomics features or conventional PET-metrics for predicting lymph node involvement (LNI; A,B), extracapsular extension (ECE; C,D), and Gleason score (GS; E,F) in the training dataset or combined validation cohort.
Training dataset AUC is the mean cross-validation AUC.
Table 4.
The AUC, sensitivity and specificity of the radiomics-based models per outcome.
Fig 2.
The effect of Combat harmonization in the external validation dataset on model performance per machine learning classifier for lymph node involvement (LNI), extracapsular extension (ECE), and Gleason score (GS) prediction.
Fig 3.
The effect of adding clinical parameters to radiomics data on model performance for the combined validation dataset.
Fig 4.
ROC-curves of the baseline models in the training dataset and the combined validation dataset.
The pre-radical prostatectomy MSKCC-nomogram for lymph node involvement (LNI; A) and extracapsular extension (ECE; B) prediction and the biopsy baseline model for postoperative Gleason score (GS; C) prediction.