Nuklearmedizin 2020; 59(02): 96
DOI: 10.1055/s-0040-1708140
Leuchttürme
Leuchtturm-Sitzung 7: TechnoRadiomics
© Georg Thieme Verlag KG Stuttgart · New York

in vivo D’Amico score for low-high risk and biochemical recurrence prediction in prostate patients with PET/MRI and machine learning

L Papp
1   Medical University of Vienna, Wien
,
CP Spielvogel
1   Medical University of Vienna, Wien
,
D Krajnc
1   Medical University of Vienna, Wien
,
M Grahovac
1   Medical University of Vienna, Wien
,
T Beyer
1   Medical University of Vienna, Wien
,
M Hartenbach
1   Medical University of Vienna, Wien
,
M Hacker
1   Medical University of Vienna, Wien
› Author Affiliations
Further Information

Publication History

Publication Date:
08 April 2020 (online)

 

Ziel/Aim PET/MRI is a promising modality to characterize prostate lesions. To date, the pre-operative diagnosis of prostate cancer is performed by the D’Amico score, utilizing biopsy-based Gleason scoring. We aimed to establish an in vivo D’Amico score for post-operative and biochemical risk prediction built on PET/MRI and machine learning (ML).

Methodik/Methods 74 dual-tracer (18F-FMC+68Ga-PSMAHBED-CC) multi-parametric prostate PET/MRI cases were involved in our study. Gleason-annotated, full-mount histopathological slices guided the PET/MRI lesion delineation (Hermes Hybidref 3D). For each lesion, 56 radiomics features were extracted from PET/MRI as of [1]. Ensemble learning [2] was utilized to build lesion-specific low (=G4) and benign (=G3) risk probability predictors (LH and BM respectively). The summed LH and BM prediction probabilities of the two largest lesions were mapped into +1, +2 and +3 ML risk scores. The in vivo D’Amico score was composed of the ML, the original PSA and the clinical stage risk scores in each patient. 1000-fold Monte Carlo (MC) cross-validation estimated the performance of the in vivo D’Amico score to predict post-operative low vs mid-high risk (IPR) and biochemical recurrence (IBCR) no-yes scores respectively.

Ergebnisse/Results The IPR prediction performance was SENS 83 %, SPEC 88 % and ACC 84 % (reference D’Amico SENS 74 %, SPEC 94 % and ACC 80 %). The IBCR performance values were SENS 100 %, SPEC 81 % and ACC 86 % (reference D’Amico SENS 78 %, SPEC 65 % and ACC 69 %).

Schlussfolgerungen/Conclusions Our results indicate that the in vivo D’Amico score can accurately predict low vs mid-high post-operative risk as well as biochemical recurrence of prostate patients as the alternative of the biopsy-based D’Amico score.

 

  • Literatur/References

  • 1 Papp L. , et al: Optimized feature extraction for radiomics analysis of 18F-FDG-PET imaging. JNM 2018. (in press).
    Google Scholar
  • 2 Papp L. et al: Glioma Survival Prediction with Combined Analysis of In Vivo 11C-MET PET Features, Ex Vivo Features, and Patient Features by Supervised Machine Learning. JNM 2018 ( (59) ): 892-899