Abstract:Background and objective: Cervical cell classification has important clinical significance in cervical cancer screening at early stages. In contrast with the conventional classification methods which depend on hand-crafted or engineered features, Convolutional Neural Network (CNN) generally classifies cervical cells via learned deep features. However, the latent correlations of images may be ignored during CNN feature learning and thus influence the representation ability of CNN features. Methods: We propose a novel cervical cell classification method based on Graph Convolutional Network (GCN). It aims to explore the potential relationship of cervical cell images for improving the classification performance. The CNN features of all the cervical cell images are firstly clustered and the intrinsic relationships of images can be preliminarily revealed through the clustering. To further capture the underlying correlations existed among clusters, a graph structure is constructed. GCN is then applied to propagate the node dependencies and thus yield the relation-aware feature representation. The GCN features are finally incorporated to enhance the discriminative ability of CNN features. Results: Experiments on the public cervical cell image dataset SIPaKMeD from International Conference on Image Processing in 2018 demonstrate the feasibility and effectiveness of the proposed method. In addition, we introduce a large-scale Motic liquid-based cytology image dataset which provides the large amount of data, some novel cell types with important clinical significance and staining difference and thus presents a great challenge for cervical cell classification. We evaluate the proposed method under two conditions of the consistent staining and different staining. Experimental results show our method outperforms the existing state-of-arts methods according to the quantitative metrics (i.e. accuracy, sensitivity, specificity, F-measure and confusion matrices). Conclusions: The intrinsic relationship exploration of cervical cells contributes significant improvements to the cervical cell classification. The relation-aware features generated by GCN effectively strengthens the representational power of CNN features. The proposed method can achieve the better classification performance and also can be potentially used in automatic screening system of cervical cytology. (C) 2020 Elsevier B.V. All rights reserved.

Cervical Histopathology Image Clustering Using Graph Based Unsupervised Learning

A Cervical Histopathology Image Clustering Approach Using Graph Based Features

Clustering of Cervical Histopathology Images Based on Minimum Spanning Tree

Cervical Histopathology Image Classification Using Multilayer Hidden Conditional Random Fields and Weakly Supervised Learning

Supervised Graph Hashing for Histopathology Image Retrieval and Classification

Whole Slide Cervical Cancer Classification Via Graph Attention Networks and Contrastive Learning

Cervical Histopathology Image Classification Using Ensembled Transfer Learning

Cervical Cell Classification with Graph Convolutional Network.

Cervical Nuclei Segmentation in Whole Slide Histopathology Images Using Convolution Neural Network

GraphCL: Graph-based Clustering for Semi-Supervised Medical Image Segmentation

CCF-GNN: A Unified Model Aggregating Appearance, Microenvironment, and Topology for Pathology Image Classification

Whole Slide Cervical Cancer Screening Using Graph Attention Network and Supervised Contrastive Learning

Graph-Based Medical Image Clustering

Full-automatic Computer Aided System for Stem Cell Clustering Using Content-Based Microscopic Image Analysis

Automatic cell nuclei segmentation and classification of cervical Pap smear images

A Cervical Histopathology Dataset for Computer Aided Diagnosis of Precancerous Lesions

Clinical applications of graph neural networks in computational histopathology: A review

Application of Graph-Curvature Features in Computer-Aided Diagnosis for Histopathological Image Identification of Gastric Cancer

Mine Local Homogeneous Representation by Interaction Information Clustering with Unsupervised Learning in Histopathology Images

Histopathology Whole Slide Image Analysis with Heterogeneous Graph Representation Learning

Optimize Deep Learning Models for Prediction of Gene Mutations Using Unsupervised Clustering