Segmentation of Squamous Columnar Junction on VIA Images using U-Net Architecture

Abstract

Cervical cancer is the second most common cancer that affects women, especially in developing countries including Indonesia. Cervical cancer is a type of cancer found in the cervix, precisely in the Squamous Columnar Junction (SCJ). Early screening for cervical cancer can reduce the risk of cervical cancer. One of the popular screening tool methods for the detection of cervical pre-cancer is in the Visual Inspection with Acetic Acid (VIA) method. This is due to the level of effectiveness, convenience, and low cost. VIA examination is done by applying 3-5% acetic acid to the cervical area. After applying acetic acid, a lesion called Acetowhite (AW) will be seen. AW is a precancerous lesion surrounding the SCJ. Early screening of the SCJ region will facilitate the detection of AW lesions in future studies. This method detect and segment unstructured and small patterns even though the amount of data is limited with good accuracy results. This paper proposes a method for the detection and segmentation of the SCJ region on VIA images using U-Net. This study is the first research conducted using the CNN method to perform segmentation tasks in the SCJ region. The proposed method is applied to nine different models. By using number of filter variations such as default, downfilter and upfilter and post processing methods Fix Threshold and Otsu Threshold are tested to find the best performance results. The best performance results was achieved by U-Net upfilter architecture with 90.86%, 56.5%, 75.69%, 34.09%, 41.24%, and 56.91% for Pixel Accuracy, Mean IoU, Mean Accuracy, Dice coefficient, Precision, and Sensitivity respectively. It is hoped that in the future, an easy, inexpensive, and effective automatic cervical pre-cancer detection device can be developed and accessed widely.