Transfer Learning approach for Classification of Cervical Cancer based on Histopathological Images

Cervical Cancer is the fourth most prominent type of cancer in which occurrence and death rates are steadily rising, particularly in developing nations of the world. Cervical adenocarcinoma (CADC) and cervical squamous cell carcinoma (CSCC) are two distinct subtypes of cervical cancer (CC) with different genetic profiles, risk factors, and clinical behaviors. An essential benchmark for the diagnosis of cervical cancer subtypes (CADC and CSCC) is hematoxylin and eosin (H& E) histopathological whole-slide images (WSIs). Incorrect identification and misclassification frequently arise from the uniform appearance of pathological cervical images, the substantial number of interpretations, the lengthy and tedious process of reading, and the pathologist’s inadequate work experience. To overcome these limitations, computer-aided diagnosis (CAD) technologies, such as traditional machine learning (ML) and deep learning (DL), have recently have been employed to identify patterns that are helpful for medical diagnosis. In previous years, transfer learning (TL) achieved tremendous achievements in the field of DL, and the use of TL technology in cervical histopathology categorization became an emerging research area. The objective of this research is to create an automated, comprehensive system for identifying cervical cancer subtypes (CADC and CSCC) from histopathology images using TL. This study used 59 high-resolution WSIs, 45 WSIs were used to train the model, and 14 were used for testing. The proposed model obtained an accuracy of 85% and an AUC score of 86%.