DETECTION OF CERVICAL CANCER FROM UTERINE CERVIX IMAGES USING TRANSFER LEARNING ARCHITECTURES

Abstract

Cervical cancer is a common and serious cancer affecting more than half a million women worldwide. For cervical cancer disease management, prognosis prediction, or optimizing medical intervention, early detection of the disease is critical. It is one of the types of cancer that can be successfully treated, as long as it is diagnosed early and managed effectively. In this study, an image processing-based solution was proposed for the diagnosis of cervical cancer from uterine cervix images using transfer learning architectures to reduce the workload and assist the experts. The proposed transfer learning model was tested using a publicly available dataset, which includes 917 uterine cervix images. Uterine cervix images were enhanced and brightness level using the histogram equalization method and denoised using the Gaussian filter. Then, the performances of AlexNet, DenseNet201, MobilenetV2, Resnet50, Xception, and VGG19 transfer learning architectures were compared. The transfer learning model performance was evaluated using the 10-fold cross-validation method. VGG19 transfer learning algorithm had the highest performance. VGG19 transfer learning algorithm achieved 98.26% accuracy, 0.9671 f1-measure, 0.9896 specificity, 0.9631 sensitivity, 0.9711 precision, 0.9552 Matthews correlation coefficient (MCC), and 0.955 kappa statistic. The combination of histogram equalization, Gaussian filter, and the VGG19 transfer learning approach can be used for accurate and efficient detection of cervical cancer from uterine cervix images. In this study, more accuracy was achieved compared to the known related studies in the literature.