Cervical cancer is the fourth most commonly occurring cancer in women. It can be prevented by regular screenings to find precancerous cells through the Papanicolaou Smear Test and microscopy which can show cellular abnormalities. However, the manual microscopic screening for the nuclear abnormalities is subjective and prone to error, making automated detection a necessity. This study aims to quantify the nuclear features related to shape characteristics of normal and abnormal cells from pap smear images and examine potential detection of multi classes. Using the ground truth images of normal and abnormal cells we extracted the nuclear shape features that corresponded to the classified cells such as normal and three categories of abnormal: mild, moderate and severe; that is four classes. The dataset of the nuclear shape features were visually plotted as a heat map and bubble plots using the ground truth or known predetermined labelled normal, mild-, moderate- or severe abnormal cells, and also without any such labeling. By clustering, 78 - 89% of the cells were successfully matched with the ground truth. Further, we found that more than 4 classes were obtained. In conclusion, by data visualization techniques we can classify precancerous cells.
Fernandez, Caroline; Patel, Tanvi; and Pandya, Krushang
"Analysis of Human Cervical Cell Images from Pap Smears for Classification,"
Undergraduate Research: Vol. 2:
1, Article 6.
Available at: https://kb.gcsu.edu/undergraduateresearch/vol2/iss1/6