Exemplar pyramid deep feature extraction based cervical cancer image classification model using pap-smear images. (March 2022)

• Record Type: Journal Article
• Authors: Yaman, Orhan; Tuncer, Turker
• Is Part Of: Biomedical signal processing and control. Volume 73(2022)
• Publication Date: 2022-03
• Subjects: Cervical cancer -- Pap-smear datasets -- Classification -- Exemplar Pyramid Deep Feature Extraction -- DarkNet; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2021.103428
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: Automated classification of the pap-smear image using exemplar pyramid deep feature extraction technique. A shallow classifier(Cubic SVM) with high classification ability. Two cervical cancer datasets (SIPaKMeD and Mendeley LBC) are used for pap-smear image classification. Obtained 98.26% and 99.47% accuracies for SIPaKMeD and Mendeley LBC datasets respectively. Abstract: Cervical cancer is a common type of cancer in women worldwide. Detection of this type of cancer in the early stages is very important for the treatment process. Early diagnosis/detection is very important for the treatment of cervical cancer. The golden standard of diagnosing cervical cancer is the pap-smear test. To automatically diagnose cervical cancer, machine learning is a good solution and many computer vision/deep learning-based models have been presented in the literature. In this study, an exemplar pyramid deep feature extraction-based method has been proposed for the detection of cervical cancer. The prime purpose of our proposal is to classify cervical cells in pap-smear images for the detection of cancer. SIPaKMeD and Mendeley Liquid Based Cytology (LBC) datasets have been used to develop our exemplar pyramid deep feature generator. The phases/steps of the proposed exemplar pyramid structure-based model are; (i) transfer learning-based feature extraction using DarkNet19 or DarkNet53 networks in an exemplar pyramid structure and the proposed feature generator creates 21, 000 features. By deploying Neighborhood Component Analysis (NCA), the most informative/weighted 1000 features from the generated 21, 000 features. The selected 1000 features by NCA are classified with the Support Vector Machine (SVM) algorithm. Both 5-fold cross-validation and hold-out validation (80:20) have been utilized as validation techniques. The best accuracies for the SIPaKMeD and Mendeley LBC datasets have been computed as 98.26% and 99.47%, respectively. The obtained results illustrate that the proposed exemplar pyramid model is successful to diagnose cervical cancer using pap-smear images.