PT - JOURNAL ARTICLE AU - Tobias Ryden AU - Martijn van Essen AU - Ida Marin AU - Johanna Svensson AU - Peter Bernhardt TI - Deep learning generation of synthetic intermediate projections improves <sup>177</sup>Lu SPECT images reconstructed with sparsely acquired projections AID - 10.2967/jnumed.120.245548 DP - 2020 Aug 01 TA - Journal of Nuclear Medicine PG - jnumed.120.245548 4099 - http://jnm.snmjournals.org/content/early/2020/08/27/jnumed.120.245548.short 4100 - http://jnm.snmjournals.org/content/early/2020/08/27/jnumed.120.245548.full AB - The aims were to decrease 177Lu-SPECT (single-photon emission computed tomography) acquisition time by reducing the number of projections and to circumvent image degradation by adding deep learning–generated synthesized projections. Methods: We constructed a deep convolutional U-structured network for generating synthetic intermediate projections (CUSIP). The number of SPECT investigations was 352 for training, 37 for validation, and 15 for testing. The input was every fourth projection of 120 acquired SPECT projections, i.e., 30 projections. The output was 30 synthetic intermediate projections (SIPs) per CUSIP. SPECT images were reconstructed with 120 or 30 projections, or 120 projections where 90 SIPs were generated from the 30 projections (30-120SIP); using 3 CUSIPs. The reconstructions were performed with two ordered subset expectation maximization (OSEM) algorithms: attenuation-corrected (AC)-OSEM, and attenuation, scatter, and collimator response–corrected (ASCC)-OSEM. Image quality of SIPs and SPECT images were quantitatively evaluated with root mean square error, peak signal-to-noise-ratio (PSNR), and structural similarity (SSIM) index metrics. From a Jaszczak SPECT Phantom, the recovery and signal-to-noise ratio (SNR) were determined. In addition, an experienced observer qualitatively assessed the SPECT image quality of the test set. Kidney activity concentrations, as determined from the different SPECT images, were compared. Results: The generated SIPs had a mean SSIM value of 0.926 (0.061). For AC-OSEM, the reconstruction with 30-120SIP had higher SSIM (0.993 vs. 0.989; p&lt;0.001) and PSNR (49.5 vs. 47.2; p&lt;0.001) values than the reconstruction with 30 projections. ASCC-OSEM had higher SSIM and PSNR values than AC-OSEM (p&lt;0.001). There was a minor loss in recovery for the 30-120SIP set, but SNR was clearly improved compared to the 30-projection set. The observer assessed 27/30 of the images reconstructed with 30 projections as having unacceptable noise levels, whereas corresponding values were 2/60 for the 30-120SIP and 120 projection sets. Image quality did not differ significantly between the 30-120SIP and 120 projection reconstructions. The kidney activity concentration was similar between the different projection sets, excepting a minor reduction of 2.5% for the ASCC-OSEM 30-120SIP. Conclusion: Adopting synthetic intermediate projections for sparsely acquired projections considerably recovers image quality and could allow reduced SPECT acquisition time in clinical dosimetry protocols.