Cellular Outline Segmentation using Fractal Estimators

SALVATELLI A., CAROPRESI J., DELRIEUX C., IZAGUIRRE M. F., Y CASCO V.

Journal of Computer Science and Technology

EDULP

Año: 2007 vol. 7 p. 14 - 22

1666-6046

Segmentation in biological images is essential forthe determination of biological parameters that allowthe construction of models of several biologicalproblems. This helps to establish clear relationshipsbetween those models and the parameter estimation,and for elaboration of key experiments that givesupport to biological theories. Segmentation is theprocess of qualitative or quantitative information extraction(shape, texture, physical and geometric properties,among others). These quantities are neededto compute the biological descriptors for furtherclassification (v.g., cell counting, development stageassessment, and many others). This process is almostalways supervised (i.e., human assisted), since thequality of the images that are produced with classicmicroscopy technologies have defects that in generaldisallow the application of unsupervised segmentationtechniques.In this paper we investigate the use of the a localfractal dimension estimation as an image descriptorfor microscopy images. This local descriptor appearsto be robust enough to perform unsupervisedor semisupervised segmentations, specifically in ourstudy. We applied this technique on microscopyimages of amphibian embryos’ skin in which, usingimmunofluorescence techniques, we have labeledthe cell adhesion molecule E-Cadherin.This molecule is one of the key factors of the Ca2+-dependent cell—cell adhesion. Segmentation of thecellular outlines was performed using a processingworkflow, which can be repeatedly applied to a set ofsimilar images, from which information is extractedfor characterization and eventual quantification purposes.