Conditional Super Resolution Generative Adversarial Network for Cervical Cell Image Enhancement

Abstract

Abnormal cellular development in the cervix causes cervical cancer. Undergoing Pap smear screening & HPV tests at regular intervals possibly helps identify the disease at an early stage, aiding diagnosis and treatment. Pathologists are involved in the screening to detect abnormal cells in the slide. Removing a portion of the slide with poor quality could improve nuclei detection by enabling better visualization of some important information. In this study, a Conditional Generative Adversarial Network (CGAN) with a Super Resolution technique is applied for cervical cell images to generate photo-realistic images. Poor-quality images are not removed; rather, they are converted into good-quality images. It ensures equal distribution of image generation in terms of classes in a given dataset. The performance of the recommended technique is measured using PSNR (Peak Signal-to-Noise Ratio), SSIM (Structural Similarity)and Inception Score(IS). The proposed methodology outperforms other augmentation techniques and Variational Autoencoders (VAE) based algorithms.