Open Access

ARTICLE

Cervical Cancer Classification Using Combined Machine Learning and Deep Learning Approach

Hiam Alquran^1,2, Wan Azani Mustafa^3,4,*, Isam Abu Qasmieh², Yasmeen Mohd Yacob^3,4, Mohammed Alsalatie⁵, Yazan Al-Issa⁶, Ali Mohammad Alqudah²

1 Department of Biomedical Engineering, Jordan University of Science and Technology, Irbid, 22110, Jordan
2 Department of Biomedical Systems and Informatics Engineering, Yarmouk University 556, Irbid, 21163, Jordan
3 Faculty of Electrical Engineering Technology, Campus Pauh Putra, Universiti Malaysia Perlis, 02000, Arau, Perlis, Malaysia
4 Advanced Computing, Centre of Excellence (CoE), Universiti Malaysia Perlis (UniMAP), 02000, Arau, Perlis, Malaysia
5 The Institute of Biomedical Technology, King Hussein Medical Center, Royal Jordanian Medical Service, Amman, 11855, Jordan
6 Department of Computer Engineering, Yarmouk University, Irbid, 21163, Jordan

* Corresponding Author: Wan Azani Mustafa. Email:

Computers, Materials & Continua 2022, 72(3), 5117-5134. https://doi.org/10.32604/cmc.2022.025692

Received 01 December 2021; Accepted 02 March 2022; Issue published 21 April 2022

Abstract

Cervical cancer is screened by pap smear methodology for detection and classification purposes. Pap smear images of the cervical region are employed to detect and classify the abnormality of cervical tissues. In this paper, we proposed the first system that it ables to classify the pap smear images into a seven classes problem. Pap smear images are exploited to design a computer-aided diagnoses system to classify the abnormality in cervical images cells. Automated features that have been extracted using ResNet101 are employed to discriminate seven classes of images in Support Vector Machine (SVM) classifier. The success of this proposed system in distinguishing between the levels of normal cases with 100% accuracy and 100% sensitivity. On top of that, it can distinguish between normal and abnormal cases with an accuracy of 100%. The high level of abnormality is then studied and classified with a high accuracy. On the other hand, the low level of abnormality is studied separately and classified into two classes, mild and moderate dysplasia, with ∼ 92% accuracy. The proposed system is a built-in cascading manner with five models of polynomial (SVM) classifier. The overall accuracy in training for all cases is 100%, while the overall test for all seven classes is around 92% in the test phase and overall accuracy reaches 97.3%. The proposed system facilitates the process of detection and classification of cervical cells in pap smear images and leads to early diagnosis of cervical cancer, which may lead to an increase in the survival rate in women.

---

Keywords

---