TY - JOUR T1 - Automated characterisation of PSMA-PET/CT uptake patterns of bones in prostate cancer JF - Journal of Nuclear Medicine JO - J Nucl Med SP - 1183 LP - 1183 VL - 60 IS - supplement 1 AU - Robert Seifert AU - Aaron Scherzinger AU - Florian Büther AU - Martin Boegemann AU - Matthias Weckesser AU - Xiaoyi Jiang AU - Michael Schäfers AU - Kambiz Rahbar Y1 - 2019/05/01 UR - http://jnm.snmjournals.org/content/60/supplement_1/1183.abstract N2 - 1183Objectives: Bone metastases of prostate cancer are often associated with a poor clinical outcome. Therefore, the concise characterisation of bone involvement is clinically demanded for patients with prostate cancer. PSMA (Prostate-specific membrane antigen) targeted PET/CT imaging has emerged as standard tool for whole body staging of prostate cancer. We propose a novel workflow for an automated analysis of the PSMA uptake patterns in bones by PET texture analysis. Additionally, non-textural PET and textural CT features were calculated for comparison. It is hypothesised that texture analysis of all bone and not single metastases is favourable for the detection and the determination of the extent of bone metastases. Methods: 95 68Ga-PSMA-11 PET/CT acquisitions of prostate cancer patients with and without bone metastases were included in this retrospective clinical case series. A consensus reading on the presence and extent of bone metastases was performed by two nuclear medicine specialists, which served as reference for the workflow evaluation. Both for PET and CT, fourteen textural image features were calculated and compared to conventional PET metrics (like SUVmax or SUVmean). Results: The best PSMA PET texture analysis feature (‘standard deviation’) had a high sensitivity (93 %) and specificity (96 %) for bone metastases detection (AUC = 0.97). Best non-textural image feature was the Mean Molecular Volume of bone metastases (AUC = 0.98, sensitivity = 90 %, specificity = 100 %). CT texture analysis obtained only a sensitivity of 74 % and specificity of 72 % (‘relative standard deviation’, inverted ROC: AUC = 0.75). The non-textural image feature met-bone (PET derived fraction of metastatically affected bone) correlated best with the visual extent characterization (rho = 0.94). Conclusions: The proposed automated bone analysis workflow is detecting bone metastases with high sensitivity and specificity through textural features. For bone metastases detection and extent characterisation, CT texture analysis is inferior to PET derived image features. The proposed PET analysis workflow might serve as a reference tool for the detection and monitoring of prostate cancer bone involvement in future studies. ER -