Adaptive parallel interference cancellation (PIC) has been recently introduced to the signal detection of multiple-input multiple-output (MIMO) systems. However, it suffers from error propagation when operated in time-varying channels. In this paper, an adaptive two-stage PIC with the minimum variance (MV) criterion is proposed to solve the problem. Adaptation with the MV criterion is realized with a decision feedback generalized sidelobe canceller (DFGSC). In the first-stage cancellation, a special structure involving dual DFGSCs is developed. AU adaptation operations are implemented with the least-mean-square (LMS) algorithm. Simulation results show that the proposed two-stage PIC detection can significantly outperform the conventional adaptive PIC detection in fast time-varying channel environments.