Table 1.

Image preprocessing techniques are presented, with their rationale and advantages.

Image Preprocessing Technique	Rationale	Advantage
Normalization	MRI data contain arbitrary intensity units and grey-level intensity that can be homogenized with intensity outlier filtering (e.g., calculating the mean and standard deviation of grey levels and excluding those outside a definite range such as mean ± 3 times the standard variation).	Reducing the heterogeneity due to varying pixel grey-level value distribution across exams
Resampling	Images with different spatial resolutions can be uniformed and either upscaled or downscaled to isotropic voxel spacing.	Increases reproducibility by making texture features rotationally invariant
Discretization	Grouping pixels into bins based on intensity ranges, which is conceptually similar to creating a histogram.	A greater number of bins (or a smaller bin width) tend to preserve image details at the cost of noise. Conversely, noise reduction can be achieved by reducing the number of bins (or increasing bin width) but will cause the image to lose detail.
Bias field correction	MRI can suffer from spatial signal variation caused by the magnetic field being intrinsically inhomogeneous.	Correct undesired inhomogeneities
Image filtering	Application of edge enhancing (e.g., Laplacian of Gaussian) or decomposition (e.g., wavelet transform) filters to obtain additional image volumes from which to extract features.	May emphasize useful image characteristics while reducing noise