TY - JOUR T1 - <strong>Automated detection and quantification of prostatic PSMA uptake in SPECT/CT using a deep learning algorithm for segmentation of pelvic anatomy</strong> JF - Journal of Nuclear Medicine JO - J Nucl Med SP - 30 LP - 30 VL - 59 IS - supplement 1 AU - Karl Sjostrand AU - Aseem Anand AU - Jens Richter AU - Kerstin Johnsson AU - Konrad Gjertsson AU - Lars Edenbrandt AU - Vivien Wong Y1 - 2018/05/01 UR - http://jnm.snmjournals.org/content/59/supplement_1/30.abstract N2 - 30Objectives: 99mTc MIP-1404 (1404), a prostate-specific membrane antigen (PSMA) targeted imaging agent is currently under investigation for the detection of clinically significant disease in prostate cancer. Quantitative assessment of tracer uptake in SPECT/CT images requires substantial user interaction and introduces observer variability. Our objective was to develop a deep learning algorithm for automated detection and quantification of prostatic 1404 uptake in SPECT/CT images in a clinical setting. Methods: We developed a deep learning algorithm based on convolutional neural networks for automatically segmenting the prostate and pelvic bones from CT images. The algorithm was designed to process both high- and low-dose CT images as well as whole and part body field of views, with no manual interaction necessary. The training material consisted of 100 diagnostic CT images (all male) with complete and distinct segmentations, performed manually, for relevant anatomical regions. Validation of the algorithm was performed using SPECT/CT images from 102 high-risk prostate cancer patients in a phase 2 clinical study (MIP-1404-201, NCT01667536) who underwent 1404 imaging prior to radical prostatectomy. These scans were previously quantified manually using the OsiriX medical image viewer (Pixmeo SARL), by measuring the maximum uptake in a circular ROI placed inside the prostate in the slice and region with highest uptake values determined visually. The automated algorithm uses its volumetric segmentations to measure uptake at every voxel in the prostate, and registers the maximum uptake. The Pearson correlation coefficient was used to assess the concordance between manual and automated quantification of prostatic uptake. Results: The deep learning algorithm based on the 100 cases training material had 2.7 million parameters and was optimized using a variant of gradient descent. In the test set, 1404 images of 34 patients (33%; 34/102) were excluded due to excessive CT artifacts, incomplete data and/or data format problems. Computation time on the evaluable patients (N=68) was 13 seconds (per case) on commodity hardware. The automated maximum uptake value was significantly correlated to the manually obtained in the prostate (r=0.95, 95% CI=[0.91,0.97], p&lt;0.0001; slope=0.89, 95%CI=[0.80,0.98]), supporting validation of the algorithm. Conclusions: The feasibility of a fully automated and fast algorithm for quantification of 1404 uptake in the prostate has been demonstrated. The potential clinical application of this and related measures will be further explored in prospective clinical studies. ER -