Experiments in Large Ensemble for Segmentation and Classification of Cervical Cancer Biopsy Images.




The task of classifying cells within a sample tissue from a patient into the classes of normal and cancer requires tissue segmentation. After segmentation one of the important extractable features to discriminate between normal and cancer cervical cells is their mean nuclear volume. Due to the rapid reproduction of cancer cells, they have higher mean nuclear volume than typical normal cells. In this paper we propose a large ensemble of segmentations which discriminates between normal and cancer cases based on the single feature of mean nuclear volume. It uses four basic segmentors and generates new segmentations from their results and then extracts the mean nuclear volume from these new segmentations. The used dataset contains multiple images from hematoxylin and eosin (H&E) stained archival tissue sections from 29 normal and 32 cancer patients. Results show a clear separation between the two classes in the dataset. These findings support the use of automatic stereology in a staging of cervical tissue.

Please Contact Us for more information about this work.