Quantification of the efficiency of segmentation methods on medical images by means of non-euclidean distances

PASTORE, J.; BALLARIN, V.; MOLER, E.

Journal of Physics

American Institute of Physics

Lugar: St Louis, MO, USA ; Año: 2007 p. 1 - 9

1742-6588

To quantify the efficiency of a segmentation method, it is necessary to do some validation experiments, consisting generally in comparing the result obtained against the expected result. The most direct method for validation is the comparison of a simple visual inspection between the automatic segmentation and a segmentation obtained manually by a specialist, but this method does not guarantee robustness. This work presents a new similarity parameter between a segmented object and a control object, that combines a measurement of spatial similarity through the Hausdorff metrics and the difference in the contour areas based on the symmetric difference between sets.