Diffusion-based molecular communications has becoming a promising approach for end-to-end communication between nano-machines. However, due to the randomness nature of molecule diffusions, the arriving time of each molecule is hard to predict and the late coming molecules will become a source of interference when doing detection. This effect will cause a severe degeneration of the system performance. Through our explorations, the conventional methods for mitigating the interference cannot be directly applied in these diffusion-based molecular communication environments. One of the main reasons is that the channel response is now time-variant. In this paper, we first apply Bayesian criterion to design an end-to-end communication system using quantity-based modulation (i.e. The information is imbedded in the number of molecules transmitted). Inspired by the idea of decision feedback in traditional communications, we propose an inter-symbol interference (ISI) cancellation approach. We also demonstrate the cancellation process based on our proposed quantity-based modulation system. Numerical results show that the proposed ISI cancellation method can help us achieve a reliable end-to-end transmission between nano-scale devices.