TY - JOUR T1 - miPSMA Index: Comprehensive and Automated Quantification of <sup>18</sup>F-DCFPyL (PyL-PSMA) PET/CT for Prostate Cancer Staging JF - Journal of Nuclear Medicine JO - J Nucl Med SP - 1435 LP - 1435 VL - 61 IS - supplement 1 AU - Kerstin Johnsson AU - Hannicka Sahlstedt AU - Johan Brynolfsson AU - Johnny Dang AU - John Ceccoli AU - Tess Lin AU - Jens Richter AU - Karl Sjostrand AU - Lars Edenbrandt AU - Matthias Eiber AU - Aseem Anand Y1 - 2020/05/01 UR - http://jnm.snmjournals.org/content/61/supplement_1/1435.abstract N2 - 1435Objectives: For clinical trials as well as clinical practice, precise and reproducible measurements of disease state based on imaging are of great value. PROMISE (Eiber et al. 2018) defines a standardized language for reporting PSMA PET/CT reads, and groups patients into categories based on visual scores of PSMA expression, miPSMA scores. In this study, we propose miPSMA index, a modified methodology, to automate and quantify PSMA expression at lesion level and at patient level in men with prostate cancer. Methods: Each bone, lymph or prostate lesion is quantified with a Lesion miPSMA index (LPI), a continuous adaptation of the miPSMA score. An LPI equal to 1 implies that the mean lesion standard uptake value (SUVmean) equals blood pool reference uptake, LPI=2 implies that SUVmean equals liver reference uptake, and LPI=3 implies that SUVmean is equal to or above twice the liver reference uptake. For each tissue type the individual LPI’s are aggregated into the patient-wide miPSMA index: PSMA-weighted Total Lesion Volume (PTLV). For example, all bone LPI’s are multiplied with the respective lesion volume and aggregated into the PSMA-weighted Total Lesion Volume (PTLVbone). PTLVlymph and PTLVprostate are computed analogously. Here we report two levels of miPSMA index, LPI and PTLV, for bone/lymph node/prostate on 194/97/43 PET/CT scans from the PyL Research Access Program. We report PTLV grouped by indication: Screening (S), Newly Diagnosed (ND), Recurrent or Suspected Recurrence (R) and Metastatic (M). In the automated computation, relevant organs are segmented in the CT with deep learning and projected into PET space. An algorithm then detects hotspots in the PET image and labels them based on the anatomical segmentation. Hotspots considered to represent lesions are manually selected and adjusted if needed; also lesions not detected by the algorithm are segmented. Reference SUV’s are computed from deep learning segmentations of the thoracic part of aorta and the liver in the CT that are transferred to PET and eroded to account for PET/CT misalignment. The reference values used are the mean of the values the interquartile range in each segment. Results: In the data set, 94% of the LPI’s were between 1 and 3, with minimum LPI for bone/lymph/prostate being 0.82/1.1/1.3. Comparison between indications is based on the mean of all patient PTLV’s within the interquartile range of patients with the indication, hence excluding outlying patients.This gives an ordering of PTLV for the indications which is S&lt;ND&lt;R&lt;M for bone, S=M&lt;R&lt;ND for lymph node and M&lt;R&lt;S&lt;ND for prostate. The automated hotspot detection algorithm demonstrated a sensitivity of 92.1% for bone lesions (97.2% in the 52 automatically segmented bones) and 96.2% for regional lymph node lesions, with on average 17/23 bone/lymph hotspots detected per scan. In a sample of 20 PET/CT scans, the correlation (Pearson’s r) between automated and manual reference was 0.92 for blood and 0.87 for liver. The blood and liver automated reference values are based on a larger image volume compared to the manual reference method and are hence expected to be more robust. Conclusions: We have developed the automated miPSMA index on a lesion level (LPI) and a patient level (PTLV), for comprehensive quantification of PSMA expression in prostate, lymph node and bone lesions of patients with prostate cancer. Further studies will evaluate the clinical relevance of the miPSMA index as an imaging biomarker for staging, prognosis and assessment of treatment response in comparison to traditional indices such as Total Lesion Volume and Total Lesion Uptake. Figure 1: PSMA-Weighted Total Lesion Volume (PTLV). Figure 2: Lesions PSMA Index (LPI) for all annotated lesions in the data set. Figure 3: Manual versus automatic reference values for blood pool and liver. ER -