Pix2Pix Generative Adversarial Network for Cellular Nuclei and Cytoplasm Segmentation on Pap Smear Images

In medical imaging for Pap smear tests, accurately identifying regions of interest, such as the nucleus and cytoplasm, remains a critical challenge due to the complex morphology and overlapping structures in cervical cell images. This complexity increases the risk of misidentification, potentially leading to false positives in computer-assisted diagnosis. To address this issue, this study introduces a novel approach by developing and evaluating a framework for the precise segmentation of nuclei and cytoplasm in cervical cell images using a cGAN-based model, Pix2Pix, applied to a dataset validated by specialists. The generated images are compared with target images, converted to binary, and an AND operation is performed to evaluate pixel overlap in the areas of interest. The evaluation metrics highlight a segmentation accuracy of 88.8 % and sensitivity of 89.62 % for nuclei, while for cytoplasm, precision reached 89.62 % and sensitivity 99.34 %. The Jaccard indices were 80.89 % for nuclei and 96.71 % for cytoplasm. These results demonstrate the effectiveness of the model in segmenting nuclei and cytoplasm in cervical cells.