Anisotropic Regularization of Posterior Probability Maps Using Vector Space Projections
M. A. RodriguezFlorido, R. Cardenes, C.F. Westin, C. AlberolaLopez, J. RuizAlzola
Computer Aided Systems Theory (EUROCAST'03), Lecture Notes in Computer Science 2809
Pages 597606
February 2428, 2003
Abstract
In this paper we address the problem of regularized data classification. To this extent we propose to regularize spatially the classposterior probability maps, to be used by a MAP classification rule, by applying a noniterative anisotropic filter to each of the classposterior maps. Since the filter cannot guarantee that the smoothed maps preserve their probabilities meaning (i.e., probabilities must be in the range [0, 1] and the classprobabilities must sum up to one), we project the smoothed maps onto a probability subspace. Promising results are presented for synthetic and real MRI datasets.

Figure 5.a shows a zoomed area of a slice from the original MRI volume, Fig. 5.b shows the MAP classification (the parameters are provided from a groundtruth segmentation) and Fig. 5.c shows the regularized MAP segmentation using the same probabilistic characterization. The labels for the segmentation are white for WM, gray for GM, dark gray for CSF, and black for the background. Notice how the regularized MAP provides much more spatially coherent classifications while preserving the borders. 
Reference
RodriguezFlorido MA, Cardenes R, Westin CF, AlberolaLopez C, RuizAlzola J. Anisotropic regularization of posterior probability maps using vector space projections. In RM Diaz, AQ Arencibia, eds., Computer Aided Systems Theory (EUROCAST'03), Lecture Notes in Computer Science 2809. Las Palmas de Gran Canaria, Spain: Springer Verlag, 2003;597606.Bibtex entry
@INPROCEEDINGS{rodriguezfloridoEUROCAST03, author = {M. A. RodriguezFlorido and R. Cardenes and C.F. Westin and C. AlberolaLopez and J. RuizAlzola}, title = {Anisotropic Regularization of Posterior Probability Maps Using Vector Space Projections}, editor = {Roberto Moreno Diaz and Alexis Quesada Arencibia}, booktitle = {Computer Aided Systems Theory (EUROCAST'03), Lecture Notes in Computer Science 2809}, pages = {597606}, month = {February 2428}, year = {2003}, address = {Las Palmas de Gran Canaria, Spain}, publisher = {Springer Verlag}, note = {} }
Grants
NIH P41RR13218 (NAC), CIMIT, TIC20013808C02Research areas
DTMRI, TensorCopyright Information
SpringerVerlag