FCM-UNet: a cervical cell segmentation model with fine-grained feature extraction capability

Cervical cell screening is an effective means of early detecting cervical cancer. The shape, size, color and texture of the nucleus of cervical cells are of great significance in determining whether the cells have undergone pathological changes. To address the issues of inaccurate segmentation of contour details in cell nuclei and the overall low segmentation precision, a novel model named FCM-UNet is proposed, which is an enhancement of the conventional U-Net. Firstly, a Fine-grained feature squeeze (FGFS) module is embedded in U-Net to extract more fine-grained semantic features, which can portray small changes and complex morphology of nucleus edges. Secondly, a Convolutional block attention module (CBAM) is introduced between the encoder and decoder to guide the model focusing on effective feature regions, enhancing the feature representation ability. Finally, a Multi-scale feature extraction (MFE) module based on spatial pyramid pooling is constructed to enable the model to acquire significant multi-scale feature information. The segmentation precision and Dice similarity coefficient of FCM-UNet on the Herlev dataset are 96.75% and 94.15%, respectively. Compared with other state-of-the-art algorithms, FCM-UNet demonstrates superiority in processing and recognizing the contour details of cervical cell, which can provide more accurate and reliable support for cervical cell image analysis and cervical cell screening.