A fast model independent method for automatic correction of intensity nonuniformity in MRI data

Elizabeth A. Vokurka, Neil A. Thacker, Alan Jackson
1999 Journal of Magnetic Resonance Imaging  
A novel non-parametric approach for correcting intensity non-uniformity in magnetic resonance (MR) data in an image volume is described. This model is based on smooth shifts in intensity within homogeneous materials and does not require radio frequency (RF) coil, tissue class models, or optimisation. The advantage of this computationally fast method is that it can be applied early in quantitative analysis while being independent of pulse sequence and insensitive to pathological processes. This
more » ... lgorithm has been tested on both real and simulated data. Application to tissue segmentation and functional MR imaging has shown a marked improvement in quantitative analysis. BACKGROUND Historically, the emphasis on the development of magnetic resonance imaging (MRI) has been on the production of easily interpreted, high contrast, noise free diagnostic images. Recently, the medical physics community has begun to examine the extent to which MRI can provide quantitative information to investigate physiology and structure. The motivation for the non-uniformity correction algorithm presented below has been an interest in the use of multi-spectral MR data for quantitative analysis and measurement of tumour tissue. Most tissues are relatively homogeneous and should display a high degree of intensity uniformity throughout a given volume. This offers the potential of automatic and accurate structure determination for a wide variety of clinical tasks. However, in the interest of obtaining low noise, high contrast images, some MR scanners often make undocumented adjustments to the amplification and digitisation of the RF image signal. These hardware adjustments can vary between the z-plane (phase encode) slices of a volume image and between scans for a particular patient. A similar problem exists for within slice variation due to sensitivity effects, particularly for surface receiver. All magnetic resonance images suffer from signal intensity non-uniformity to some extent causing a smooth drift of average intensity across constant regions of tissue. Radio frequency (RF) coil asymmetry, gradient-driven eddy currents, pulse sequence parameters, and overall patient anatomy all potentially contribute to this gain drift [3]. Many simple analysis techniques, requiring repeatable measurements, are impossible or impeded due to changing MR scanner calibration and patient and coil positioning over time. This has potential implications across areas of MR image analysis, including perfusion measurements, functional imaging, segmentation, and co-registration. Some analysis techniques are designed to cope with these effects, but often the approach is to ignore the problem and assume that consequent errors are small. While average intensity non-uniformities of 10-30% in standard head coil data will affect visual diagnosis minimally, the accuracy of quantitative procedures such as co-registration, segmentation, and functional imaging can be significantly affected [4] [8] . It is essential to develop a fast, robust method of compensating for unmodelled gain changes across an image volume, reducing systematic error in quantitative measurements.
doi:10.1002/(sici)1522-2586(199910)10:4<550::aid-jmri8>3.0.co;2-q pmid:10508322 fatcat:yg4udnqswvgspbplkjmq6xxbmi