Multiclass Classification of Cervical Pap Smear Images Using Deep Learning-Based Model

Abstract

One of the primary reasons for women's deaths is cervical cancer. The most important procedures that are to be performed to ensure the reduction of cervical cancer's side effects are possible diagnosis and the most definitive possible medical treatment. One of the best methods for identifying this type of cancer is using PaP smear images. For cervical cancer detection in PaP smear images, this research proposes a novel hybrid deep learning approach. The proposed methodology for cervical cancer classification performs the four efficient stages. A shape-based iterative method is used to detect nuclei in cell segmentation, and a marker-control watershed approach is used to separate overlapping cytoplasm. From the regions of segmented nuclei and cytoplasm, the practical features are extracted in the features extraction step. The simulated annealing integrated with a wrapper filter is employed for efficient feature selection. The classification of cervical cancer from pap-smear images is achieved using an attention-based nested classification network (Anu- Net) based on deep learning. The SIPaKMeD dataset is used for experiment analysis. The experimental results reveal that the developed deep-learning network model enabled high classification accuracy. The accuracy for a binary class problem was 99.95%; for a threeclass problem was 99.98%; and a five-class problem was 99.74%. The proposed approach significantly outperformed existing deep learning models in binary class, three-class, and five-class problems than the existing approaches.