Symmetric diffeomorphic registration for multi-modal brain MRI

• Autores: Jesús Omar Ocegueda González
• Directores de la Tesis: Mariano J. J. Rivera Meraz (dir. tes.)
• Lectura: En la Centro de Investigación en Matemáticas (CIMAT) ( México ) en 2016
• Idioma: inglés
• Enlaces
  • Tesis en acceso abierto en: cimat.repositorioinstitucional.mx
• Resumen

  • This thesis is dedicated to the study of computational methods for registration of magnetic resonance images (MRI) of the brain. The problem of brain MRI registration appears as a sub-problem of several image processing tasks in which a correspondence between anatomical points of at least two brain MR images needs to be established. Of particular interest is the problem of magnetic susceptibility-induced geometric distortions in echo-planar images.

    Among the contributions of this thesis, I present a new matching functional for multi-modal brain MRI registration. The relationship between both image modalities is modeled as a global non-linear function that makes the local intensity correspondence as linear as possible. The proposed matching functional results from a maximum-likelihood estimation of our model’s parameters and is implemented to drive the Symmetric Normalization algorithm to obtain inverse-consistent diffeomorphic transforms. Experimental results using manually annotated brain MR images show that our similarity measure consistently outperforms traditional multi-modal metrics such as Mutual Information. Additionally, we present advances on the application of the proposed technique for correction of susceptibility-induced geometric distortions.