Adapted MRF Segmentation of Multiple Sclerosis Lesions Using Local Contextual Information.
Nagesh K. Subbanna, Simon J. Francis, Doina Precup, D. Louis Collins, Douglas Lorne Arnold, Tal Arbel · MIUA · 2011
We present a fully automated technique to segment lesions from multimodal brain MRIs of patients with Multiple Sclerosis. We describe an adapted Markov Random Field that uses intensity at every voxel, its neighbourhood intensity difference information and neighbouring voxel class information to infer voxel labels at every voxel. We test our technique on 25 real, clinical MS volumes evaluated by five experts. Our method outperforms two state of the art methods: one an outlier based MRF technique and the other a hybrid Bayesian-MRF technique both qualitatively and according to the Dice similarity coefficients and the number of present negative lesions.