A MultiScale Algorithm for Nuclei Extraction in Pap Smear Images

Abstract

This work presents a new automated method which manages multiscale information and combines segmentation and classification algorithms for nuclei extraction in pap smear images. The accuracy of the segmentation algorithms was evaluated using the comparison functions relative distance error and object consistency error. The harmonic mean of sensitivity and specificity was used in the classification evaluation. The evaluation of different alternatives shows as the best result the combination of the Shape Detection and Artificial Neural Network. The multiscale approach provides a convenient way to combine information from different resolutions. It outperforms the usual algorithms because there is no single “true” scale for a Pap smear images. The proposal is fast enough and accurate and, so, it is very helpful for cell screening. Usually, the algorithms that include as one of their steps the classification of information, do not justify the choice made. On this work a study is included on which is the best classification method for the Nuclei Extraction in Pap Smear Images.