Detection of cervical cancer cells based on strong feature CNN-SVM network

• Published in: Neurocomputing (Amsterdam) Vol. 411; pp. 112 - 127
• Main Authors: Dongyao Jia, A., Zhengyi Li, B., Chuanwang Zhang, C.
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 21.10.2020
• Subjects: Cancer cell detection; CNN-SVM; Sample amplification; Strong features; Strong features; Sample amplification; Cancer cell detection; CNN-SVM
• ISSN: 0925-2312; 1872-8286
• DOI: 10.1016/j.neucom.2020.06.006

•A dataset expansion method based on rotating transformation and scaling was proposed.•A dual channel network for feature extraction and classification was constructed and designed the overall network architecture. Convolutional neural network is used to extract abstract features from data sets while strong feature path extracts strong features based on prior knowledge. After feature dimensionality reduction and feature fusion, these two features are input into the SVM classifier to complete the classification of cervical cells.•GLCM+Gabor with proper fusion ratio and SVM were applied after experiment.•The proposed strong feature CNN-SVM classifier out performed than other classification methods. Traditional screening of cervical cells largely depends on the experience of pathologists, which also has the problem of low accuracy and poor efficiency. Medical image processing combining deep learning and machine learning shows its superiority in the field of cell classification. A new framework based on strong feature Convolutional Neural Networks (CNN)-Support Vector Machine (SVM) model was proposed to accurately classify the cervical cells. A method fusing the strong features extracted by Gray-Level Co-occurrence Matrix (GLCM) and Gabor with abstract features from the hidden layers of CNN was conducted, meanwhile the fused ones were input into the SVM for classification. An effective dataset amplification method was designed to improve the robustness of the model. The proposed method was evaluated on two independent datasets with the metrics of accuracy (Acc), sensitivity (Sn), and specificity (Sp). Our approach outperformed than the state-of-the-art models with the Acc, Sn, and Sp of 99.3, 98.9, 99.4 for 2-class detection in the mass, respectively. The results indicated that the strong feature CNN-SVM model could be applied in cell classification for the early screening of cervical cancer.