An Enhanced Fuzzy Deep Learning (IFDL) Model for Pap‐Smear Cell Image Classification
S. Rakesh, Smrita Barua, D. Anitha Kumari, E. Naresh · 2024
The conventional method of categorizing cervical cancer types relies heavily on the expertise of pathologists, which is associated with a lower degree of precision. The utilization of colposcopy is an essential element in the prevention of cervical cancer. Colposcopy has been a crucial component in the reduction of cervical cancer frequency and humanity rates over the past five decades, in conjunction with precancer screening and treatment. The rise in workload has resulted in reduced diagnostic efficiency and misdiagnosis during vision screening. The utilization of the convolutional neural network (CNN) model in medical image processing has demonstrated its superior performance in the cervical cancer type within the realm of cavernous learning. The present study puts forth two convolutional neural network architectures based on deep learning for the identification of cervical cancer through the analysis of colposcopy images. The models employed in this research are VGG19 (TL) and Colposcopy Ensemble Network (CYENET). The utilization of VGG19 as a transfer learning approach has been implemented in the CNN architecture for research purposes. The Colposcopy Ensemble Network has been developed as a novel model for the automatic classification of cervical cancers from colposcopy images. The model's precision, selectivity, and responsiveness are evaluated. The VGG19 model exhibited a classification accuracy of 70.3%. The outcomes for VGG19 (TL) are moderately satisfactory. The kappa score analysis of the VGG-19 perfect inferred that the model falls within the moderate classification category. The findings of the experiment indicate that the CYENET model demonstrated noteworthy levels of sensitivity, specificity, and kappa scores, specifically, 90.4%, 95.2%, and 88%, correspondingly. The CYENET model exhibits an enhanced classification accuracy of 90.1%, surpassing the VGG19 (TL) model by 10%.