TY - JOUR T1 - Semi-automated quantification of osseous prostate metastatic burden to the spine and pelvis with sodium fluoride PET/CT JF - Journal of Nuclear Medicine JO - J Nucl Med SP - 1307 LP - 1307 VL - 58 IS - supplement 1 AU - Matthew Thorpe AU - Rustain Morgan AU - Timothy Turkington AU - Jorge Oldan AU - Aaron Nelson AU - Bennett Chin Y1 - 2017/05/01 UR - http://jnm.snmjournals.org/content/58/supplement_1/1307.abstract N2 - 1307Objectives: Quantification of the burden of osseous metastases via 18F-fluoride PET/CT predicts prognosis in patients with prostate cancer. Rapid, reliable automation of tumor burden measurement would facilitate multi-center therapeutic trials and clinical interpretations in a busy clinical environment. Whole body single threshold (ST) segmentation is rapid and reproducible, but may have reduced accuracy in the face of variability across patients, institutions and regions of the skeleton. We previously reported a patient specific regional threshold approach (RT) for rib metastases detects more lesions than an ST model, without loss in specificity. We now investigate semi-automated characterization of osseous metastases to the spine and pelvis.Methods: 18F-sodium fluoride (NaF) PET/CT scans (n=40 scans nested within 36 patients) were analyzed using customized workflows in MIM 6.4 with MIM Encore software. Consensus reads by two radiologists established true positives and degenerative changes resulting from increased 18F-NaF activity. Skeleton was reliably resolved from soft tissue using an SUV > 1.5 threshold; however this threshold also included bladder, ureters and renal calyces. The user separated the GU tract from bone and the spine from the ribs, skull, and pelvis in a 40 second manual segmentation step. The workflow then applied 1) a global single SUV threshold > 10 SUV units, and 2) a regional threshold RT defined as mean SUV in a user identified normal vertebral body multiplied by a regional factor based on prior reports of normal relative regional uptake. Free response ROC curves were generated against consensus reads as the gold standard. Segmented regions identified as degenerative changes on consensus reads were counted as false positives to compare specificity. Area under the adjusted free response ROC curve (JAFROC AUC) was compared, as were true positive, false positive and false negative rates.Results: Consensus reads revealed 15 normal scans and 25 scans containing 169 osseous metastases (range of number of metastases 0-51, IQR 1-5). There was no difference in true positive fraction, false negative fraction or overall area under the FROC curve; however false positives were significantly decreased using a relative threshold (p<0.05, see Table).Conclusion: The use of patient and region specific semi-automated segmentation thresholds decreased the false positive rate without decreasing the true positive rate. The high prevalence of degenerative changes in the spine and pelvis still required substantial user oversight to remove false positive NaF activity for accurate estimates of lesion number; efficient semi-automated methods for identification of degenerative change are needed. Future efforts will attempt to address these issues via coregistration of SUV and Hounsfield units, as well as discriminating center of activity within cortex vs. medulla to improve resolution of metastatic from degenerative disease. Research Support: N/A View this table: ER -