In this paper, we propose a phase noise cancellation algorithm in MIMO-OFDM systems where each transmit (TX) and receive (RX) antenna is equipped with independent oscillator. Most of previous works have considered phase noise model only in SISO-OFDM or simplified phase noise model in MIMO systems assuming that each TX and RX end shares one local oscillator. Recently, however, the local oscillator for each TX and RX antenna can be different for some reasons, in particular for the case of mmWave based wireless systems. In our paper, maximum a posteriori (MAP) estimator is used to jointly estimate phase noise at TX and RX and detect data symbols. Also we derive Bayesian Cramér-Rao bound (BCRLB) for the phase noise processes. At the end of this paper, we demonstrate from our simulation results that the application of proposed algorithm improves the bit-error-rate performance and its mean squared error can be close to BCRLB.