Automated Cell Nuclei Segmentation in Overlapping Cervical Images Using Deep Learning Model

Recently, an automated cell nuclei segmentation is an attractive research area in the cervical cancer image analysis with the overlapping cells. Due to the poor contrast, diverse cells and the occlusion, the segmentation suffers from several issues such as inaccurate region detection and the lack of boundary refinement. To alleviate such issues, this paper proposes the deep learning framework with the boundary refinement techniques. Prior to boundary refinement, this paper utilizes the noise removal technique called Neighborhood Concentric Filtering (NCF) based on connected component analysis. Then, the proposed work segments the cell nuclei and cytoplasm with clear boundary refinement. The proposed work utilizes the N-ary ternarybased texture pattern extraction to collect the features that describe the regions clearly. Then, this paper utilizes the fisher model to select the relevant features in order to reduce the dimensionality that directly reduces the time complexity. Finally, the Deep Learning (DL) models through the neural network approaches classifies the abnormal cells effectively. Besides, this paper investigates the effectiveness in terms of the various performance parameters over the existing methods in cervical cancer image analysis for earlier diagnosis applications.