The correlation and Wiener filter techniques introduced by Sijbers effectively improved the SNR and CNR of the MR brain images. The greatest SNR increase was produced by the Gaussian filter; this filter also caused the greatest degradation of image resolution, however. The large effect on SNR is most likely due to the large neighborhood size of the Gaussian filter over space. While the median filter used a 2x2 neighborhood and the average filter a 3x3, the Gaussian filter covered a 5x5 neighborhood and thus had a much greater area over which to average. Of course, this then resulted in the significant blurring over this large neighborhood that led to the large drop in resolution. Although the loss of resolution number is an estimate, it definitely shows the inverse relationship between SNR improvement and resolution loss in MR imaging. It is perhaps most interesting to compare the images from the noisy image, gaussian filter, and contrast filter. Even though the contrast filter lacks even the CNR apparent in the original noisy image, its resolution has improved, and when compared to the resolution loss produced by the gaussian filter, the improvement is remarkable. Overall, the average filter may have provided the best effect (with the highest SNR/resolution loss ratio as a marker); but it is important to remember that at some point the resolution loss will outweigh any SNR improvements. Blurring of the image to increase SNR may drown out lesions or tumors that would otherwise have been detected. Consequently, when designing filters to improve SNR or CNR in MR images, the concomitant effect on resolution loss must also be considered.

If this project were to be extended, I would suggest looking at alternate methods of SNR estimation to test the reliability of this method. Also, one could try using different neighborhood dimensions in the filters I implemented to find a size that provides excellent SNR without blurring the image and decreasing the resolution significantly. Additionally, further filters can be developed in an attempt to improve the existing filters.