Dr. Art White and students David Diller and Mark Willis worked on a cell classification research project to detect abnormalities in cervical epithelial cells. Cervical epithelial cells may be classified as normal or abnormal by statistically analyzing morphological cell characteristics obtained through image processing. After heuristically determining six pertinent cell characteristics, the Cover Learning Integer Linear Programming (CLILP) machine learning algorithm, developed by K. Cios (University of Toledo), was applied to the statistical data obtained from a training database of 273 cervical cells classified by a trained cytotechnologist.
The CLILP algorithm determined which cell features and values best discriminated normal from abnormal cells. The team then generated rules based on those discriminating features determined by CLILP. These rules were then applied to a test set of 135 cervical cells and performance was evaluated based upon comparison of the automated system versus cell classification by the cytotechnologist.
The system agreed with the cytotech in 81 percent of the classifications. In 16 percent of the cases, the system labeled cells “abnormal” where the cytotech considered them “normal.” Only 3 percent of the cells were labeled “normal” when the cytotech judged them “abnormal.”
The results of this research were used in conjuction with a project to detect the location of cell nuclei to decrease the amount of time it takes a cytologist to analyze a series of Pap smear slides.
Need more information about how you can be a part of our challenging learning environment? Send us an email, and we’ll send you a CSE Info Kit and a Taylor University T-shirt.