We demonstrated that our novel landmark-based registration pipeline outperforms both TPS and the underlying nonlinear intensity-based registration without landmark usage. This highlights the potential of automatic landmark correspondence detection for improvement of lung CT registration accuracy.