Abstract–The jPET-D4 is a novel brain PET scanner which aims to achieve not only high spatial resolution but also high scanner sensitivity by measuring 4-layer depth-of-interaction (DOI) information. In this work, we present software strategies for 3D image reconstruction and imaging performance of the jPET-D4 prototype. The dimensions of a system matrix for the jPET-D4 become 4 billion (coincidence pairs) x 5 million (image elements) when a 25cm diameter FOV is sampled by a 1.53mm3 voxel. The size of the system matrix is estimated at 142peta (P) byte with the accuracy of 8 byte per element. The on-the-fly calculation is usually used to deal with a huge system matrix. However we can not avoid the extension of calculation time when we improve the accuracy of system modeling. In this work, we proposed an alternative approach based on the pre-calculation of the system matrix. The 142P byte system matrix was compressed into 13.4GB by (1) reducing zero elements, (2) applying the 3D-expanded DOI compression method, (3) factorizing with respect to ring differences and (4) restricting the maximum ring difference to 54 (with only 10% loss of the number of LORs). Histogram-based 3D OSEM based on geometrical system modeling was implemented. After evaluating basic imaging performance though phantom experiments, a normal volunteer was scanned and the first human brain images were obtained.