Ashish Raj1, XIaobo Shen2, Thanh D. Nguyen, PhD3, Susan A. Gauthier4
1Department of Radiology, Weill Cornell Medical Collge, New York, NY, United States; 2Department of Computer Science, Cornell University, Ithaca, NY, United States; 3Department of Radiology, Weill Cornell Medical College, New York, NY, United States; 4Department of Neurology and Neuroscience, Weill Cornell Medical College, New York, NY, United States
To introduce a new post-processing technique for quantifying myelin in the brain, with specific application to the examination of demyelinating disesaes like Multiple Sclerosis. T2 Relaxometry is a popular MRI technique which can separate the contribution of various tissue components in the brain, thereby quantifying the myelin content of brain regions. It works by capturing several MRI scans at different echo times, followed by a numerical fitting procedure to fit multiple components exponentially relaxing at different T2 time constants. Unfortunately, the post-processing required to obtain myelin maps from T2 data is a hard numerical problem due to ill-posedness of the problem. Consequently, the T2 distributions and the resulting myelin water fraction (MWF) maps become very sensitive to noise and are frequently difficult to interpret diagnostically. Hence T2 relaxometry typically necessitates very high SNR T2 scans which can take several hours for whole brain coverage clearly a clinically unfeasible proposition. Here, we propose a new way of solving the inverse problem in T2 relaxometry by imposing spatial smoothness constraints and by restricting the relaxing T2 distribution to 2 Gaussian-shaped peaks corresponding to myelin water and intra/extra-cellular water. The method greatly improves robustness to noise, reduces spatial variations and definition of white matter fiber bundles in the brain. This allows it to be used on fast but low-SNR spiral acquisitions which take only 10 minutes for whole brain coverage.