Profile likelihood maximizes over $\psi$, while Marginal likelihood integrates over $\psi$.

Marginal likelihood is only used in frequentist inference.

Profile likelihood requires a prior distribution.

Marginal likelihood is generally computationally cheaper than Profile likelihood.