The fast-ion distribution function in fusion plasmas can only be measured indirectly by solving an ill-posed inverse problem. The inversion being ill-posed necessitates regularisation of the problem to ensure that the reconstruction of the fast-ion distribution function depends smoothly on the measurements obtained by fast-ion diagnostics. In turn, the resulting reconstruction depends on the choice of regularisation, and it is therefore beneficial to choose a physics-informed prior as regularisation scheme. In this work, we reconstruct the high-energy tail in the MeV-range of the fast-deuterium distribution in JET discharges heated by waves in the ion cyclotron range of frequencies (ICRF) using neutron and gamma-ray emission spectroscopy. We do this by applying a physics-informed prior based on collision physics and a newly formulated ICRF-physics prior, and we compare these results with numerical simulations and inversions based on a standard Tikhonov regularisation scheme. Our findings suggest that the physics-informed regularisation scheme including the ICRF prior improves the reconstructions compared with standard Tikhonov regularisation. Finally, it is shown that constraining the reconstruction to have negative gradients in the directions of phase space dictated by ICRF physics results in a reconstruction that well resembles expectations based on ICRF physics theory and numerical simulations.