Blind Dereverberation for MRI Acoustic Noise

Abstract. In Magnetic Resonance Imaging (MRI) room, strong acoustic noise during the scan not only creates harmfully annoying environment but also interferes with brain activation response. The presence of reverberation in the MRI environment interferes with efficient noise canceling techniques. In this paper, we present a blind dereverberation method for estimating the broadband MRI acoustic noise signal from the sensor measurements. The method is based on modeling the acoustic noise as an autoregressive (AR) process and using a single input multiple output (SIMO) linear prediction system. In our experiments, we use not only synthetic data but also actual recording of the reverberant MRI acoustic noise signals in a room and in a test-bed mimicking the MRI bore. A criteria based on spectral characteristics is employed to analyze the performance of this method. Experimental results show effective removal of the reverberation from the original noise signal over a broad band of signal spectrum. Results in this paper will provide further help to improve noise cancellation performance in MRI environment.