The present paper deals with the blind deconvolution problem of multiple-input multiple output finite impulse response (MIMO-FIR) channels driven by source signals which are white in the sense of second-order statistics and spatially higher-order uncorrelated but temporally higher-order colored signals. Our goal is to solve the blind deconvolution problem using a deflation approach. To this goal, we propose an algorithm obtained by modifying the super-exponential deflation algorithm (SEDA) proposed by Inouye and Tanebe [8] to the case of the blind deconvolution problem of MIMO-FIR channels driven by the source signals which possess higher-order auto-correlations, which is referred to as a modified super-exponential deflation algorithm (MSEDA). Using the MSEDA, the colored source signals which are the inputs of the MIMO-FIR channels are extracted from the outputs of the MIMO-FIR channels one by one. To show the validity of the proposed approach, some simulation results are presented.