An Improved Deep Convolutional Model for Segmentation of Nucleus and Cytoplasm from Pap Stained Cell Images

For the early detection of cervical dysplasia, automated cervical cell analysis system requires an accurate segmentation of nucleus and cytoplasm from cells. The segmentation of cellular materials from pap stained cytology image is open issue due to touching and crowded cells, presence of inflammatory cells, mucus and blood in the image. In this paper, for detecting and analyzing cell components from cervical smears, we developed a deep convolution framework using FC-Densenet56. Here images from Herlev dataset are trained and tested in deep architectures. A combination of FC-DenseNet56 and ResNet101 were used in proposed method to get an accurate result. For the comparison purpose, the results of proposed segmentation were evaluated with Precision and Dice coefficient, that achieves better results than the works reported in the literature. The performance parameters such as Precision and Dice coefficient is obtained greater than 90% and Recall and IoU got values greater than 85%. Besides cervical smear images, the proposed methodology can be adopted for segmentation of other cytology images.