A Two stage approach for Detection of Invasive and Cervical Intra-Epithelial Neoplasia using Machine Learning and Image Processing Methodologies

Representing around 6.6% of the cancer cases in women, Cervical Cancer is the fourth most common cancer in women. Cancer is a pernicious disease marked by rapid multiplication and growth of malignant cells in the body. The threat that it poses is evident from the fact that it is the second leading cause of death worldwide and consumes about 1 in 6 individuals worldwide. Of all the types of cancer, Cervical Cancer is the eighth most occurring cancer. Cancer causing infections, such as hepatitis and Human Papilloma Virus (HPV), are accountable for nearly 25% of cancer instances in low- and middle-income countries. Diagnosed in around 122,844 women in India and the cause of death for 67,477 women, this morbid disease can be best tackled when there is early diagnosis. It occurs when there is an aberrant growth of the cells of the cervix which further infect other tissues of the body. The lower part of the uterus that connects it with vagina is the cervix. The abnormal growth of cells mentioned earlier is caused by Human Papilloma Virus (HPV) infection, a sexually transmitted infection.The infection spreads in three stages of Cervical Intra-epithelial Neoplasia (CIN), and finally the most severe stage results in the onset of Cervical Cancer. This paper aims at implementing various Machine Learning Methodologies for first detecting the likelihood of transmission of HPV infection, the leading cause of Cervical Cancer by using a questionnaire involving questions related to the sexual activity of the individuals. Later, it aims at classifying the stage of Cervical Intra-Epithelial Neoplasia which can help in diagnosis and early treatment of the disease, to avert the onset of Cervical Cancer. For achieving this, we use a set of classifiers on the personal and medical detail provided by the user for predicting the likelihood of onset of cervical cancer in stage 1. In stage 2, image processing techniques are used to obtain features which are then given to the classifier to classify them into precancerous stages.