Methods and Theory : Congealing for Bias Removal


Congealing was developed with the idea of reducing the variability of images under spatial transformations. However, the same idea can be used to reduce the variability or remove "noise" with other types of globally coherent transformation. In this work, we explored the idea of removing unwanted, low-frequency perturbations of magnetic resonance images (MRI) using the idea of joint image alignment.

The images below are pseudo-color MRI's of the brains from 20 different babies. The brightness of many of the images has been distorted by undesirable magnetic artifacts in the MRI. For example, the image in the third column of the third row shows a brightening trend from bottom to top. This brightening is an artifact and does not represent the condition of the tissue being imaged.

If you drag your mouse over the image, you see the results of the MRI bias-corrected images, which was done with our congealing algorithm. See the papers below for more details.



Graduate Students


  • Erik Learned-Miller and Parvez Ahammad.
    Joint MRI bias removal using entropy minimization across images.
    In Neural Information Processing Systems (NIPS) 17, pp. 761-768, 2005.
  • Erik Learned-Miller and Vidit Jain.
    Many heads are better than one: Jointly removing bias from multiple MRs using nonparametric maximum likelihood.
    In Proceedings of Information Processing in Medical Imaging, pp. 615-626, 2005.
  • Erik Learned-Miller.
    Data driven image models through continuous joint alignment.
    In IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI), 28:2, pp. 236-250, 2006.