In this letter, we investigated the optimal minimum-mean-squared-error (MMSE) based successive interference cancellation (SIC) strategy designed for lattice-reduction aided multiple-input multiple-output (MIMO) detectors. For the sake of generating the MMSE-based MIMO symbol estimate at each SIC detection stage, we model the so-called effective symbols generated with the aid of lattice-reduction as joint Gaussian distributed random variables. However, after lattice-reduction, the effective symbols become correlated and exhibit a non-zero mean. Hence, we derive the optimal MMSE SIC detector, which updates the mean and variance of the effective symbols at each SIC detection stage. As a result, the proposed detector achieves a better performance compared to its counterpart dispensing with updating the mean and variance, and performs close to the maximum likelihood detector.