Unsupervised Enhancement Method for Papanicolaou-Stained Cervical Cancer Pathology Images based on Zero-MSRCP
Zeming Zhu, Huiyan Jiang, Wenbo Pang, Xian‐Hua Han · 2024
Cervical cancer is a malignant tumor originating from the epithelial cells of the cervix and is one of the most common cancers in the female reproductive system. We propose a new cervical cell image enhancement network(ZeroDCE-MCI). Specifically, ZeroDCE-MCI takes color images as input and adjusts pixel values adaptively through a zero-reference deep estimated curve to obtain visually enhanced images. Then, a multi-channel illumination estimation module is incorporated to further adjust the overall image brightness, and optimize the network through a weighted loss function. Finally, the output image of the network is weighted fused with the enhanced images processed by the color-preserving Multi-Scale Retinex algorithm (MSRCP). We named the overall fusion method Zero-MSRCP, which can effectively enhance the pathological images of cervical cancer stained with Pap. We validate the effectiveness of this method using a publicly available dataset (SIPaKMeD) and a large tertiary hospital's cervical cancer pathology image dataset. The results show that, after applying our image enhancement method, the Peak Signal-to-Noise Ratio (PSNR) reaches 23.73, demonstrating a significant improvement compared to other methods. Our approach not only enhances the clarity of cell texture structures but also improves image quality, thereby having a significant positive impact on computer-assisted diagnosis.