Page 141 - Quantitative Imaging of Small Tumours with Positron Emission Tomography

Chapter 6 Our study has several limitations. First, the data set was relatively small. Still, the significant high cross-validated prediction scores indicate that even for such a training dataset size the ML models were able to identify high-risk patients in independent data. Enlargement of the current dataset will likely improve model stability and potentially model calibration. Finally, external model validation was not yet performed. In such analysis, harmonization of image processing and tumor delineation method is recommended (23). Conclusions [18F]PSMA PET radiomic features analyzed with ML are significantly predictive for LNI, presence of any metastasis, and high-risk pathological tumor features in primary PCa patients. These data demonstrate that the spatial distribution and levels of PSMA expression quantified on [18F]PSMA PET are related to both tumor histopathological grade and metastatic tendency. For prediction of nodal and/ or distant metastatic disease, PVC and a higher segmentation threshold seemed to improve model stability. Future multicenter external validation is needed to determine the benefits of using radiomics versus standard PET metrics. 140