First of all I have to say that the diffuse emission model used by the EGRET team is far from being perfect, which means that EGRET could have missed some sources and could see sources which do not exist. (We will know the truth soon since GLAST is already in the orbit!)
Casandjian and Grenier "diffuse emission model" is not a model but rather a template fit. They took the gas HI, H2, and "dark gas," inverse Compton, and isotropic components, sum them up and fit to the data with free normalization parameters for each component and in each of 3 energy ranges separately. In this fit, they used the gas distribution (H2, HI) in the form of 6 galactocentric rings with free normalization for each ring; this gives 15 free parameters for the diffuse emission model for each energy range. Such a fit implicitly assumes that the cosmic ray spectra (p, e) and fluxes are the same in every part of the Galaxy and thus the emissivity per H atom or per electron is the same everywhere. This is not correct, however. The cosmic ray spectra are different in different places (energy losses, diffusion, and the source distribution) and thus the emissivity is different as well. Note that the interstellar radiation field is also different in different places which affects electron energy losses and gamma-ray emission (inverse Compton). Their model, therefore, gives a wrong brightness in the places where the cosmic ray spectra differ from the average (note also the the observed diffuse emission is the line-of-sight integration of the emissivity times the gas density plus inverse Compton emission). This results in an excess or a deficit in the diffuse emission photons in different places; using this model could yield "new sources" relative to the 3EG catalog or "no detection" in the place of some of the EGRET sources. However, the diffuse emission model of Casandjian and Grenier is perhaps better than the original EGRET model.
Looking at their paper, I see that they also used one of GALPROP models of the diffuse emission with free normalizations of the components (pion decay, IC, bremsstrahlung, isotropic component, and "dark gas" component, but most of the analysis is done using the "Ring" model described above. Note that there is no "dark gas" component in the original GALPROP model and adding this component can change the background model. Besides, any background model is not free from the intrinsic systematic errors associated with the gas distribution derived from velocity information and the assumed rotation curve. For more discussion on the difficulties associated with the determination of the diffuse emission see our paper: http://adsabs.harvard.edu/abs/2007NuPhS.173...44M