The International Atomic Energy Agency (IAEA) assessed that single-photon emission computed tomography (SPECT) is one of the most attractive technique for safeguards of a spent fuel because of its intuitive evaluation capability by using a tomographic image. Even though there are over a hundred kinds of the verification technique, the IAEA suggested the need for “more sensitive and less intrusive alternatives to existing nondestructive assay instruments” for partial-defect detection. The aim of this study is to optimize a detector geometry of a SPECT system using Monte Carlo method and develop a deep-learning-based image-reconstruction-algorithm for fast verification of spent fuel assembly. The detector head consists of one-dimensional (1D) multi-slit tungsten collimator, Bismuth Germanate scintillator, and silicon photomultiplier arrays. An optimization study of the detector head was performed by evaluating the 1D projection image for Cs-137 line source. The shape of scintillator was determined by assessing the light transfer efficiency of the scintillator using DETECT2000 program. For the performance evaluation of the optimized detector, a tomographic image of twelve fuel sources in the assembly was compared with that acquired using the conventional detector developed by IAEA. A de-noised image reconstruction algorithm was developed by training the algorithm with 2040 data sets of ground truth image and FBP image obtained for different patterns of missing fuel rods in 3ⅹ3 fuel rod array. The optimized scintillator length, collimator slit width, slit length, and septal thickness were 4 cm, 0.2 cm, 5 cm, and 0.2 cm, respectively. The light transfer efficiency in the scintillator to the 3ⅹ70 mm³ and 3ⅹ3 mm² sensors were 23.40.6% and 5.30.3%, respectively. In the results of the image quality assessment, the optimized detector head shows about 1.5 times improved sensitivity, while the image quality is slightly poorer than the conventional one due to the bigger slit width. We confirmed the possibility for improving the image quality of the FBP images using the deep-learning-based image-reconstruction-algorithm. We expect that the combination of the detector head with higher sensitivity and the deep-learning-based image-reconstruction-algorithm will contribute to the faster verification of spent fuel assemblies compared to the conventional system.