Image Segmentation of Normal Pap Smear Thinprep using U-Net with Mobilenetv2 Encoder

Abstract

Pap smear is a technique to detect changes in the cells in the uterine wall. With a Pap smear, a woman can be known to have cervical cancer or not. However, the problem of cancer screening on pap smear images is largely hindered by improper cell staining and overlapping cell images. For accurate Pap smear image segmentation, this study uses the U-Net method which is better for Pap smear image segmentation. This method integrates the MobilenetV2 network and converts ordinary convolution into deep split convolution to improve transmission and feature utilization by the network, and at the same time increase the speed of feature extraction. Then the segmentation results from MobilenetV2 produce accuracy in distinguishing the nucleus, cytoplasm, and background on the pap smear image. The dataset used in this study is a normal analogue image of the Pap smear image obtained from the RepoMedUNM Database. Initial data processing is done by digitizing the image, where analog data from the Pap smear is transformed into a digital image. Based on the results of research that has been carried out, namely segmentation of Pap smear images using U-Net with MobilenetV2 encoder, the accuracy value on differences in nucleus, cytoplasm, and background cells is 98%.