Directional detection is the only admitted strategy for the unambiguous identification of galactic Dark Matter (DM) even in the presence of an irreducible background as beyond the neutrino floor. This approach requires to measure the direction of a DM-induced nuclear recoil in the keV-range. To probe such low energies, directional detectors must operate at high gain where 3D track reconstruction can be distorted by the influence of the numerous ions produced in the avalanches. We describe the interplay between electrons and ions during signal formation in a Micromegas. We introduce SimuMimac, a simulation tool dedicated to high gain detection that agrees with MIMAC measurements. We propose an analytical formula to deconvolve the ionic signal induced on the grid from any measurements, with no need of prior nor ad hoc parameter. We experimentally test and validate this deconvolution, revealing the fine structure of the primary electrons cloud and consequently leading to head-tail recognition in the keV-range. Finally, we present how this deconvolution can be used for directionality by reconstructing mono-energetic neutron spectra at 27 keV and 8 keV with an angular resolution better than 15 • . This novel approach for directionality appears as complementary to the standard one from 3D tracks reconstruction and offers redundancy for improving directional performances at high gain in the keV region.