Primordial black holes are unique probes of cosmology, general relativity, quantum gravity and non standard particle physics. They can be considered as the ultimate particle accelerator in their last (explosive) moments since they are supposed to reach, very briefly, the Planck temperature. Upper limits on the primordial black hole number density of mass $M_{star} = 5 10^{14}$ g, the Hawking mass (born in the big-bang terminating their life presently), is determined comparing their predicted cumulative $gamma$-ray emission, galaxy-wise, to the one observed by the EGRET satellite, once corrected for non thermal $gamma$-ray background emission induced by cosmic ray protons and electrons interacting with light and matter in the Milky Way. A model with free gas emissivities is used to map the Galaxy in the 100 MeV photon range, where the peak of the primordial black hole emission is expected. The best gas emissivities and additional model parameters are obtained by fitting the EGRET data and are used to derive the maximum emission of the primordial black hole of the Hawking mass, assuming that they are distributed like the dark matter in the Galactic halo. The bounds we obtain, depending on the dark matter distribution, extrapolated to the whole Universe ($Omega_{PBH}(M_{star}) = 2.4 10^{-10}$ to $2.6 10^{-9}$ are more stringent than the previous ones derived from extragalactic $gamma$-ray background and antiprotons fluxes, though less model dependent and based on more robust data. These new limits have interesting consequences on the theory of the formation of small structures in the Universe, since they are the only constraint on very small scale density fluctuations left by inflation.