9 - Machine learning approaches for acetic acid test based uterine cervix image analysis

Abstract

Visual inspection with acetic acid (VIA) is a cost-effective screening test for cancer of the uterine cervix in limited resource settings. Diluted acetic acid solution is applied on the cervix, which turns abnormal lesions white, called acetowhite (AW) lesions. There are several reasons for white patches on the cervix once acetic acid is applied, and all white patches are not necessarily due to cervical cancer. The most important deciding features for VIA positive lesions are the intensity of the acetowhitening, the margin of acetowhite lesions, and the texture within the acetowhite lesions. Identification of these features requires a considerable amount of skill. Hence, test accuracy is influenced by the expertise of the person performing the test. Further, evaluation of VIA images of the cervix suffers from high interobserver variability. Developing a decision support system that, when combined with existing screening methods, makes the screening effective and objective is desirable. Using image-processing algorithms, cervix images acquired during VIA examination can be analyzed. The focus of this chapter is to analyze these cervix images automatically to facilitate instant decision making and referral to hospitals during cervical-cancer screening.