Segmenting Anatomical Surfaces and Volumes with Markov Models
Bruce Fischl
Massachusetts General Hospital
Massachusetts General Hospital NMR Center
I present a set of techniques for embedding the physics of the imaging process that generates a class of Magnetic Resonance Images (MRI) into a segmentation and registration algorithm. This approach results in substantial invariance to acquisition parameters, as the effects of these parameters on the contrast properties of various brain structures are explicitly modeled in the segmentation process.
In addition, integrating image acquisition modeling with tissue classification allows the derivation of sequences that are optimal for segmentation purposes. Another benefit of these procedures is the generation of probabilistic models of the intrinsic tissue parameters that give rise to MR contrast (e.g., T1, proton density, T2*), thereby providing access to physiologically relevant parameters that may change with disease or demographic factors. Such changes can result in non-morphometric alterations in MR images that are otherwise difficult to detect.
Finally, we present a high-bandwidth multi-echo FLASH pulse sequence that achieves a high signal-to-noise ratio with minimal image distortion due to B0 effects. This sequence also enables explicit estimation of T2* and reduces test–retest intensity variability.
