A novel four-layered depth-of-interaction (DOI) positron emission tomography (PET) scanner is being developed at National Institute of Radiological Sciences, Japan. It aims to improve the image resolution, particularly at the edge of field of view while maintaining high sensitivity. However, inter-crystal scatter (ICS) occurs in the detector blocks of the jPET-D4. It is a phenomenon where there are multiple scintillations for a single irradiation of gamma photon due to Compton scatter in detecting crystals. Because of the Anger-type logic calculation, only one approximated position is detected by the jPET-D4 in the case of ICS. This causes error in position detection and ICS worsens the image contrast, particularly for smaller hotspot. In this paper, we propose to model an ICS probability by utilizing a Monte-Carlo simulator. It is a statistical relationship between gamma-ray first interaction crystal pair and the detected crystal pair. The ICS probability is then used to improve the system matrix of statistical image reconstruction algorithm ML-EM in order to correct the error of ICS. We have shown in computer simulation that image contrast is recovered successfully by applying the proposed method.