ESTRO 2023 - Abstract Book

S1912

Digital Posters

ESTRO 2023

Results The model returned high specificity (0.74±0.05) sensitivity (0.70±0.05), and accuracy (0.67±0.04) but suffered from low precision (0.60±0.10) on the holdout dataset, yielding an average area under the receiver operating characteristic curve (AUC) of 0.73±0.02 (see Figure 2). The lower precision is indicative of highly heterogeneous data in the small holdout set. The ensemble model achieved an AUC of 0.76, with 0.73 in sensitivity, 0.73 in specificity, 0.73 in accuracy and 0.73 in precision.

Conclusion This study illustrates that combining imaging, dosimetric and clinical features into a deep learning framework allows a robust estimation of the 2-year OS of NSCLC patients. This predictor could be used to iteratively improve the treatment planning process. Future work includes the addition of a larger external cohort to improve performance metrics, and evaluating the impact of various treatment plans on the 2-year OS.

PO-2124 Multiparametric MRI and PET to predict ISUP grade groups in prostate cancer patients

E. Nilsson 1 , K. Sandgren 1 , J. Grefve 1 , J. Jonsson 1 , J. Axelsson 1 , A. Keeratijarut Lindberg 1 , K. Söderkvist 2 , C. Thellenberg Karlsson 2 , A. Widmark 2 , L. Blomqvist 3 , S. Strandberg 4 , K. Riklund 4 , A. Bergh 4 , T. Nyholm 1 1 Umea University, Department of Radiation Sciences, Radiation Physics, Umea, Sweden; 2 Umea University, Department of Radiation Sciences, Oncology, Umea, Sweden; 3 Karolinska Institute, Department of Molecular Medicine and Surgery, Solna, Sweden; 4 Umea University, Department of Radiation Sciences, Diagnostic Radiology, Umea, Sweden Purpose or Objective To investigate the potential of multiparametric magnetic resonance imaging (mpMRI) and positron emission tomography (PET) for prostate cancer (PCa) risk stratification, using whole-mount registered histopathology as reference standard. Materials and Methods Fifty-five men planned for radical prostatectomy with elevated prostate-specific antigen and biopsy-proven PCa were included in this study. [ ⁶⁸ Ga]PSMA-11 PET/mpMRI and [ ¹¹ C]Acetate PET/CT were performed prior to surgery. Prostate glands were delineated on T2-weighted (T2w) images and a mould was 3D-printed for each patient. Resected prostate