Nuklearmedizin 2019; 58(02): 171
DOI: 10.1055/s-0039-1683670
Poster
Radiomics und Modelling
Georg Thieme Verlag KG Stuttgart · New York

Automated texture analysis of PSMA-PET/CT to characterize bone metastases and unaffected bone of patients with prostate cancer

R Seifert
1   University Hospital Münster, Department of Nuclear Medicine, Münster
,
A Scherzinger
2   University of Münster, Department of Computer Science, Münster
,
M Bögemann
3   University Hospital Münster, Department of Urology, Münster
,
M Schäfers
1   University Hospital Münster, Department of Nuclear Medicine, Münster
,
M Weckesser
1   University Hospital Münster, Department of Nuclear Medicine, Münster
,
K Rahbar
1   University Hospital Münster, Department of Nuclear Medicine, Münster
› Author Affiliations
Further Information

Publication History

Publication Date:
27 March 2019 (online)

Also available at
 
 

    Ziel/Aim:

    Prostate specific membrane antigen (PSMA) PET/CT has become a standard diagnostic procedure for whole body prostate cancer staging. The application of texture analysis to PET images is often limited to few target bone metastases. Texture analysis of all known metastases as well as of bone background might provide additional parameters for patient stratification. Here, automated PSMA-PET texture analysis is performed for all recorded bone metastases and for bone background.

    Methodik/Methods:

    A total number of 39 68Ga-PSMA-11-PET/CT acquisitions were analysed. The bone segmentations were calculated using low dose non-contrast enhanced CT acquisitions. This segmentation was used to mask all bones in PSMA-PET data. A global threshold was used to separate normal from metastatic PSMA uptake in bones, thereby segmenting bone metastases and bone background. Both for bone metastases and bone background, 15 texture analysis parameters were calculated using the Grey Level Cooccurrence Matrix and conventional statistical metrics (exemplary: relative standard deviation RSD).

    Ergebnisse/Results:

    Mean SUV and RSD of bone metastases correlated significantly with the overall extent of bone metastases (p = 0.002; p = 0.01). The same was true when unaffected bone background was analysed (p = 0.006; p = 0.02). To assess the performance of bone background for classifying patients as either lesion free or osseous metastasized, Receiver Operating Characteristics (ROC) were calculated (AUC 0.486 for mean SUV; AUC 0.135 for RSD). There were significant differences comparing bone background of patients with and without bone metastases (p < 0.001 for mean SUV, p < 0.001 for RSD).

    Schlussfolgerungen/Conclusions:

    Automated texture analysis of bone metastases revealed parameters that correlate significantly with the overall extent of bone metastases. Texture analysis of bone background correlated significantly with the overall extent of bone metastases. Future studies have to elucidate if this is due to the detection of disseminated prostate cells in the bone marrow.


    #