One of the major mechanisms associated with gas injection to recover heavy oil is the molecular diffusion of gas, which reduces the viscosity of heavy oil. Numerous efforts have been made to estimate the diffusivity of such gas in heavy oil by assuming the gas diffusivity as a constant. Meanwhile, only limited attempts have been made to determine the concentration-dependent diffusivity from pressure decay test data by using analytical, semi-analytical, or numerical methods. In this study, a generalized methodology has been developed to determine the diffusivity of gas (e.g., CO2) in a heavy oil as an exponential function of gas concentration with consideration of oil swelling. More specifically, non-intrusive experimental measurement data obtained by dynamic volume analysis (DVA) are used to compute the concentration-dependent diffusivity by integrating the finite element method (FEM) with the particle swarm optimization (PSO) algorithm. Subsequently, concentration-dependent diffusivity of CO2 from the DVA experimental data can be obtained in the form of the exponential function and then converted to the function of liquid-phase viscosity. Then, the pressure decay test data are employed to determine the optimal diffusivity among the solutions of concentration-dependent CO2 diffusivity obtained from DVA test data by using Hayduk and Cheng's formula together with the FEM. The final solution provides two coefficients for the exponential model and an exponent as well as a coefficient for Hayduk and Cheng's model. It is found that the obtained concentration-dependent diffusivity of CO2 is reasonable and accurate as well as it can be converted for use at different pressures and temperatures through the determined coefficient and exponent for Hayduk and Cheng's equation together with a linear relation between the natural logarithm of heavy oil viscosity and the dissolved gas concentration. Meanwhile, such findings in this study may be able to contribute to improve the gas injection projects such as vapor extraction or cyclic solvent injection processes.