MR image enhancement using an extended neighborhood filter

• Enhanced intensity is the weighted sum of extended neighbor intensities.
• Filter threshold and lattice size are optimized using Contrast Ratio, PSNR, and SSIM.
• Results are compared to those using Anisotropic Diffusion Filter, Wavelet Filter, and Local Histogram Equalization.
• Maximum contrast stretching for unconnected narrow foreground regions with better structural preservation.

A filtering scheme is proposed for contrast enhancement within a Region-Of-Interest (ROI) containing unconnected foreground regions of narrow spatial extent such as multiple sclerosis, or ischemic lesions in Magnetic Resonance (MR) images. This involves determination of localized multiplicative weights in the spatial domain using an extended set of neighborhood directions. The degree of enhancement is shown to depend on the number of such directions, as determined from the size of a rectangular lattice, together with a threshold value used for computing the multiplicative weights. Best performance in respect of visual quality is achieved by choosing a threshold corresponding to the maximum Contrast Ratio, and lattice size corresponding to the maximum Peak Signal-to-Noise Ratio within the ROI. It is shown that the proposed filter overrides Localized Histogram based Equalization (LHE) based techniques in terms of computational complexity, preservation of structural similarity and attaining the maximum extent of contrast stretching.