Massive Multiple-Input Multiple-Output (MIMO) has been identified as a key technology for the upcoming Fifth Generation (5G) wireless communication systems. The original Message Passing Detection (MPD) algorithm utilizes channel hardening theory to achieve good detection performance in large-scale MIMO systems. However, the computational complexity of the MPD algorithm increases with the increase of the modulation order and the number of user antennas, which is difficult to implement effectively in an actual large-scale MIMO system. The Probabilistic Approximation-MPD (PA-MPD) algorithm can reduce the computational complexity of the original MPD algorithm. In this paper, the iterative process of terminating the iterative process and updating the partial symbol probability is applied in the iterative process of the MPD algorithm, a Selective Update-MPD (SU-MPD) algorithm is proposed, which reduces the computational complexity of the MPD algorithm. The simulation results show that the computational complexity of SU-MPD algorithm can be reduced to 19% of MPD algorithm and 50% of PA-MPD algorithm under various antenna configurations, without reducing the detection performance of the algorithm.