Deep Cervical Cancer Prediction Network With Few Shots Learning for Cervical Cancer Prediction and Classification

ABSTRACT Cells that are divided into “regular” and “irregular” categories for automation‐based cervical recognition using Pap smears of Liquid Based Cytology is a very successful tumor detection technique based on cell imaging. However, precise cell segmentations are necessary for the majority of conventional classification techniques to detect cancer cells. In the existence of tissues and pathologies, accurate segmentation continues to be problematic despite 60 years of investigation in this area. Furthermore, earlier classification techniques rely solely on extracting manually created characteristics like morphology and texture. To overcome these drawbacks, the present research suggests a method that uses Cervical Cancer Prediction Network with Few Shots Learning (CCFSNet) to classify cervical tissues directly using deep learning features without applying any earlier segmentation. Here, FSNet is first pre‐trained using a dataset of cervical cells made up of image patches that have been coarsely centered on the layer of the nuclei and adaptively re‐sampled. Integration is applied to calculate the mean calculation score of a group of associated image patches during the verification stage. The Pap smear and liquid‐based cytology (LBC) datasets assess the recommended approach. Outcomes indicate that our proposed CCFSNet model is better than the earlier approaches in terms of performance accuracy with 98.5%, enhanced Specificity of 98.3%, and value of area under the curve score (AUC) of 99%. We obtained these increased performance metrics when the projected technique was employed in the Herlev benchmark Pap smear dataset and analyzed with a 5‐fold CV process. The dataset of HEMLBC can also be utilized for research to achieve similar advanced efficiency. Our approach shows promise for creating reading systems with automation support for primary cervical screening.