Efficient and safe disposal of CO2 in the subsurface depends on several conditions including a low-pressure build-up to avoid the risk of induced seismicity. CO2 storage reservoirs commonly comprise lithological heterogeneity which may contribute to pressure build-up during CO2 injection. The presence of certain rock types restricting CO2 flow, known as flow barriers, can limit the migration of CO2 leading to reservoir pressure build-up particularly near the injection point. Identifying suitable injection locations away from such rock types could require numerous high-fidelity numerical simulations due to the uncertainty of the distribution of such barriers. This study presents a new computationally efficient reservoir screening approach to minimize the probability of enhanced pressure build-up resulting from lithological heterogeneity near the CO2 injection point. The approach utilizes graph network algorithm to identify the path of the least resistance to CO2 flow between the injection location and the top of the reservoir. This path accounts for lithological heterogeneity and was coupled to the results from 50 reservoir-scale numerical simulations, capturing a range of reservoir property distributions, to derive a classification criterion for predicting grid-cell scale injectivity index. Testing showed that the approach could accurately predict the spatial variability of the injectivity index with a computational boost of up to 10,000 times.