The important thing to realize here is that neutron stars are limited to comparatively low masses. The classic mass of a "canonical" neutron star is $\sim1.4M_{\odot}$, and almost all observed examples fall into the $1.4\mathrm{-}2.0M_{\odot}$ range, with a few extremely tentative candidates approaching $2.5M_{\odot}$ (e.g. a possible $2.50\mathrm{-}2.67M_{\odot}$ compact object seen in GW190814, Abbott et al. 2020, which may or may not be a neutron star). The general relativistic bound is no higher than about $3.0M_{\odot}$, and an $18M_{\odot}$ neutron star is definitely out of the question.
With that in hand: Say GW 150914 was the result of a neutron star-black hole merger. If we take the neutron star to have the canonical mass of $1.4M_{\odot}$, the black hole would have to have a mass of $\sim3000M_{\odot}$ to produce the measured chirp mass. Even increasing the neutron star mass to $2.0M_{\odot}$ only reduces the black hole mass to a bit under $2000M_{\odot}$. It's certainly possible to have black holes in this mass range, but as the paper points out, this would lead to gravitational wave emission at a much lower frequency than was actually observed.
For comparison, the sources of the GW200105 and GW200115 detections, postulated to be neutron star-black hole mergers, had chirp masses of only $\sim3.41M_{\odot}$ and $\sim2.42M_{\odot}$, respectively, and total masses of $\sim10.8M_{\odot}$ and $\sim7.3M_{\odot}$. In both cases, the neutron stars appear to be comfortably within the observed range of neutron star masses.
