Automatic diagnosis of vulvovaginal candidiasis from Pap smear images

Abstract

Vulvovaginal candidiasis (VVC) is the most common genital infections that are seen every day in clinics. This infection is due to excessive growth of Candida that are normally present in the vagina in small numbers. Diagnosis of VVC is routinely done by direct microscopy of Pap smear samples and searching for the Candida in the Pap smear glass slides. This manual method is subjective, time consuming, labour-intensive and tedious. This study presents a computer-aided diagnostic (CAD) method to improve human diagnosis of VVC. The proposed CAD method reduces the diagnostic time and also can be worked as a second objective opinion for pathologists. Our main objective is detection and extraction of mycelium and conidium of Candida fungus from microscopic images of Pap smear samples. In this regard, the proposed method is composed of three main phases, namely preprocessing, segmentation, feature extraction and classification. At the first phase, bottom-hat filtering is used for elimination of the cervical cells and separating the background. Then decorrelation stretching and colour K-means clustering are used for Candida segmentation. Finally the extracted features used by a decision tree classifier to detect Candida from other parts of smear. The proposed method was evaluated on 200 Pap smear images and showed specificity of 99.83 and 99.62 and sensitivity of 92.18 and 94.53 for detection of mycelium and conidium, respectively.