Affiliations: [a] Department of AI&DS, Sri Shanmugha College of Engineering and Technology, Sankagiri, Salem | [b] Department of ECE, PSN College of Engineering and Technology, Tirunelveli, India
| [c] Nuclear Power Corporation of India Limited, Kudankulam, PO, Radhapuram, India
| [d] Department of CSE, Kongu Engineering College, Perundurai, Erode, India
Correspondence:
[*]
Corresponding author. A. Ahilan, Department of ECE, PSN College of Engineering and Technology, Tirunelveli, 627152, India. E-mail: [email protected].
Abstract: Cervical cancer is the most common and deadly malignancy affecting women worldwide. The prediction and treatment of this malignancy are necessary in order to avoid serious complications. In recent days, deep learning has enhanced the accuracy of cervical cancer prediction in its early stages. In this study, a deep learning based EN-FELM approach is proposed to detect and classify the cervical cells. Initially, the pap smear images are pre-processed to eliminate the background distortions. The EfficientNet is a reversed bottleneck MBConv used for feature extraction. Consequently, fuzzy extreme learning machine (FELM) is used to classify the healthy, benign, low squamous intraepithelial lesions (LSIL) and high squamous intraepithelial lesions (HSIL). The proposed model acquires the best classification accuracy on Herlev and SIPaKMeD datasets range of 99.6% and 98.5% respectively. As a result, the classification using FELM produces more efficient and accurate result which is significantly high compared to the traditional classifiers. The proposed EN-FELM improves the overall accuracy of 0.2%, 0.13% and 14.6% better than Autoencoder, LSTM and KNN with CNN respectively.
