Supervised machine learning enables non-invasive lesion characterization in primary prostate cancer with [68Ga]Ga-PSMA-11 PET/MRI

• DOI: 10.1007/s00259-020-05140-y
• Publication Date: 2020-12-20T00:07:43Z
• Authors: L. Papp; C. P. Spielvogel; B. Grubmüller; M. Grahovac; D. Krajnc; B. Ecsedi; R. A.M. Sareshgi; D. Mohamad; M. Hamboeck; I. Rausch; M. Mitterhauser; W. Wadsak; A. R. Haug; L. Kenner; P. Mazal; M. Susani; S. Hartenbach; P. Baltzer; T. H. Helbich; G. Kramer; S.F. Shariat; T. Beyer; M. Hartenbach; M. Hacker
• Keywords: Multiparametric MRI; PET-CT

Abstract Purpose Risk classification of primary prostate cancer in clinical routine is mainly based on prostate-specific antigen (PSA) levels, Gleason scores from biopsy samples, and tumor-nodes-metastasis (TNM) staging. This study aimed to investigate the diagnostic performance positron emission tomography/magnetic resonance imaging (PET/MRI) vivo models for predicting low-vs-high lesion risk (LH) as well biochemical recurrence (BCR) overall patient (OPR) with machine learning. Methods Fifty-two patients who underwent multi-parametric dual-tracer [ 18 F]FMC 68 Ga]Ga-PSMA-11 PET/MRI radical prostatectomy between 2014 2015 were included part a single-center pilot randomized prospective trial (NCT02659527). Radiomics combination ensemble learning was applied including PET, apparent diffusion coefficient, transverse relaxation time-weighted MRI scans each establish prediction model (M LH ). Furthermore, M BCR OPR predictive schemes built by combining , PSA, stage values patients. Performance evaluation established performed 1000-fold Monte Carlo (MC) cross-validation. Results additionally compared conventional standardized uptake value (SUV) analyses. The area under receiver operator characteristic curve (AUC) (0.86) higher than AUC SUV max analysis (0.80). MC cross-validation revealed 89% 91% accuracies 0.90 0.94 AUCs respectively, while standard score, TNM staging resulted 69% 70% predict respectively. Conclusion Our results demonstrate potential enhance radiomics without sampling.