Cell Segmentation in Digitized Pap Smear Images Using an Ensemble of Fully Convolutional Networks
Gergő Bogacsovics, András Hajdú, Balázs Harangi · 2021
This paper presents a method that provides reliable performance regarding cell segmentation in digitized Pap smear images. Since our final goal is the early detection of cervical cancer using scanned smear images, the proper segmentation of cells is of utmost importance. Our approach uses segmentation predictions from fully convolutional networks (FCNs) in addition to the original scanned image as its input. Our method transforms these input images to a final segmentation using a dedicated FCN architecture. Thus, our approach can be considered an ensemble-based one and outperforms state-of-the-art segmentation algorithms, achieving close to 93% accuracy and a Dice score of more than 69%.