We carry out a theoretical analysis of the uplink (UL) of a massive MIMO system with per-user channel correlation and Rician fading, using two processing approaches. Firstly, we examine the linear-minimum-mean-square-error receiver under training-based imperfect channel estimates. Secondly, we propose a statistical combining technique that is more suitable in environments with strong Line-of-Sight (LoS) components.We derive closed-form asymptotic approximations of the UL spectral efficiency (SE) attained by each combining scheme in single and multi-cell settings, as a function of the system parameters. These expressions are insightful in how different factors such as LoS propagation conditions and pilot contamination impact the overall system performance. Furthermore, they are exploited to determine the optimal number of training symbols which is shown to be of significant interest at low Rician factors. The study and numerical results substantiate that stronger LoS signals lead to better performances, and under such conditions, the statistical combining entails higher SE gains than the conventional receiver.