This paper presents a new method to segment MS lesions in multi-spectral MRI exams. Our approach leverages the speed of a GPU level set algorithm and a new data term to perform multiple segmentations of brain white matter in each exam. Differences between these segmentations allow us to estimate the distribution of MS lesions. We evaluated our method on BrainWeb and MS Lesion Segmentation Challenge 2008 data. It achieved results comparable to, or better than, two top performing algorithms. The average time required by our algorithm for Challenge08 data was 20.8 seconds.