Adapting non-local means of de-noising in intraoperative magnetic resonance imaging for brain tumor surgery

Abstract

In image-guided brain tumor surgery, intraoperative magnetic resonance imaging (iMRI) is a powerful tool for updating navigational information after brain shift, controlling the resection of brain tumors, and evaluating intraoperative complications. Low-field iMRI scans occasionally generate a lot of noise, the reason for which is yet to be determined. This noise adversely affects the neurosurgeons' interpretations. In this study, in order to improve the image quality of iMR images, we optimized and adapted an unbiased non-local means (UNLM) filter to iMR images. This noise appears to occur at a specific frequency-encoding band. In order to adapt the UNLM filter to the noise, we improved the UNLM, so that de-noising can be performed at different noise levels that occur at different frequency-encoding bands. As a result, clinical iMR images can be de-noised adequately while preserving crucial information, such as edges. The UNLM filter preserved the edges more clearly than did other classical filters attached to an anisotropic diffusion filter. In addition, UNLM de-noising can improve the signal-to-noise ratio of clinical iMR images by more than 2 times (p<0.01). Although the computational time of the UNLM processing is very long, post-processing of UNLM filter images, for which the parameters were optimized, can be performed during other MRI scans. Therefore, The UNLM filter was more effective than increasing the number of signal averages. The iMR image quality was improved without extension of the MR scanning time. UNLM de-noising in post-processing is expected to improve the diagnosability of low-field iMR images.