Student Presenter(s): Mary Margarette Sanchez, Dono Shodieva, SimranSanju Kadam, Ramses Nestor, Angel Singh
Faculty Mentor: Niharika Nath
Department: Biological and Chemical Sciences
School/College: College of Arts and Sciences, New York City

Cervical cancer screening involves pap smear staining and microscopy of the cervix cells to detect pre-cancerous abnormal nuclear morphologies. The analysis process is time-consuming and subject to human error. Quantitative analysis of cellular nuclear features may be used to increase detection efficiency and classification of normal and abnormal cells. The aim of this study is to investigate nuclear textural features and examine if these features can be used to discriminate between normal cells, moderate and severe dysplastic cells. A hospital benchmark dataset was used. We segmented the nucleus region using Cell Profiler. Angular Second Moment, Contrast, Correlation, Inverse Difference Moment, and Entropy were compared for normal versus moderate and severe dysplastic cells and K-means clustering was performed. The Contrast feature which measures the variations such as relative smoothness showed that the moderate and severe dysplastic cells had a significantly lower range (p<0.05) compared to the normal cells. Correlation, which is the measure of similarity in relative intensity values, was higher in moderate and severe dysplastic cells (p<0.001). Entropy which measures complexity was also higher in dysplastic cells. Using these features, prediction accuracies of 72-80% were obtained. Texture characteristics have the potential to discriminate between normal and dysplastic cells demonstrating potential in cervical cancer detection.