Cervical Cancer Diagnosis and AI

According to World Health Organization, cervical cancer is a leading cause of cancer whichcomes fourth in prevalence among women globally. In ۲۰۲۰, ۶۰۴ ۰۰۰ new cases and ۳۴۲ ۰۰۰deaths due to cervical cancer were reported and ۹۰% of whom were from low- and middle-incomecountries. Cervix can be divided into two parts; the endocervix which is the opening partof the cervix covered with columnar cells, and the exocervix which is the outer part, covered withsquamous epithelial cells and can be examined by the doctor. The Transformation zone is the areawhere exco- and endocervix meet and is the most common place to be cancerous. For a cell to becancerous, gradually abnormal changes needed to take place and it is categorized into differentprecancerous cells including dysplasia, cervical intraepithelial neoplasia (CIN), and squamousintraepithelial lesion (SIL). Precancerous cells are graded on a scale of ۱ to ۳ based on their abnormality.Precancerous changes can be detected by the test, Pap smear. Pap smear involves collectingcells of exco-, and to a lower extent, endocervix cells and looking for any abnormal changes inthe cells behind the microscope. Although pap smear showed promise in detecting precancerouscells in the early stages, it is an operator-based technique and requires highly skilled technicians.Moreover, the process is manual and the interpretation may be effortful and time-consuming. Toovercome these problems, machine learning and computer-based intelligence have made someprogress in cervical cell change analysis and diagnosis. By deep learning methods, images of thecervix are presented on large scale containing normal, abnormal, and artifact cells of the cervix.Given big data of cell images, an algorithm of the cell nucleus will be developed, distinguishingnormal and abnormal cells. We should first resize and adjust the Pap-slide image resolution. Theaugmentation technique should be applied to overcome the probable overfitting errors. Two subgroupsof the training and testing phases would be presented, both including normal and abnormalcells of the cervix. In total, we hope to reach the number of ۵۰۰۰ normal cells, ۳۰۰۰ atypical cells,۲۰۰۰ low-grade, and ۱۰۰۰ high-grade precancerous cells, according to similar literature availableup to now, we estimate the sensitivity and accuracy of our model to be ۹۰% or above. With ourdeveloping algorithm, we are looking forward to diagnosing false negative Pap-test results anddetecting the precancerous cells more precisely than the conventional available Pap test in ourcountry.

کلیدواژه ها:

artificial intelligence ، medicine ، cervical cancer ، pap smear ، big data