Image Processing Reference
In-Depth Information
2.2 Bias Field Reduction
It is also called intensity inhomogeneity or intensity nonuniformity, which is one of the main
problematic and challenging issues in MRI. It is a low-frequency undesirable signal that blurs
MR images and thus decreases the high-frequency contents of MRI such as contours and
Bias field is not always visible to the human eye and has a negative effect on segmentation
results. For analysis techniques such as image segmentation, the presence of intensity nonuni-
formity may reduce its reliability and accuracy dramatically, as per the hypothesis that intens-
ity is uniform in each tissue class being no longer veriied.
Thus, the correction of intensity nonuniformity can improve the application of many quant-
itative analyses.
Bias field is the smooth intensity variation in MR images caused by different factors such as:
• Nonuniform reception sensitivity
• Inhomogeneous RF
• Nonuniform reception sensitivity
Other less important parameters causing the bias field include:
• Eddy currents
• Mistuning of the RF coil
• Geometric distortion
• Patient movement
These parameters cause intensity variation over the image and the totality of the imaged ob-
ject [ 17 ] . Different methods exist to compensate for the inhomogeneity problem [ 18 - 22 ]. The
aim of this study is to show how to restore the corrupted image for MR images that corrupted
by bias field. We used two-step normalization method.
2.3 Locally Normalization Step
One of the problems, which happen in quantitative analysis of MRI, is that the results are
not comparable between consecutive scans, different anatomical regions and within the same
In order to segment MRI images in an effective approach, undesirable signals must be sup-
pressed before segmentation process. Thus, one of the major challenges is eliminating the bi-
as field. A normalization filter is required to remove low-frequency magnetic field variations
within the MR images in order to regulate image brightness and contrast while preserving de-
In our normalization method, a sliding window is applied to slide vertically on each MRI
image to compensate the current image bias field effects through histogram analysis.
In order to calculate the normalization factor, all the pixels within the region of the sliding
window are gathered as the input data. Then the standard deviation of all gathered pixel val-
ues is calculated (SD1) beside the standard deviation of the mid-line of the sliding window
(SD2). All the pixel data in the mid-line need to be shifted to the desired offset where the SD2
meets the SD1 value. At this point, the difference in the SD1 and SD2 is added to the pixel
values of the mid-line to compensate the bias field effect. When the minimum and maximum
pixel values of the mid-line are obtained, the mid-line pixel data are then stretched to its max-
imum data resolution (8-bit, 0-255) where the minimum value is zero and the maximum value
is 255 (see Equation 3 ) :
Search WWH ::

Custom Search