Artificial Intelligence-Based Support System for the Diagnosis of Cervical Lesions

Abstract

Cervical cancer affects the whole world today, being the fourth most common type of cancer diagnosed. The prevention methods applied standardized to the female population contribute to the reduction of incidence and mortality. Medical approaches include colposcopy, a way of visual inspection of the cervix, and identification of cervical, benign or malignant lesions. Being an analysis that depends entirely on the expertise of the medical professional who performs it, it can be improved by artificial methods. Deep learning has been successfully applied in the field of medical imaging and also in the classification of cervical lesions. In the presented paper, a set of convolutional networks for deep learning was used, consisting of a network with residues and a dense network, pre-trained on an extensive database and applied to a limited dataset. The assembly achieved accuracy on the validation set of 69%, improving the performance of individual models by 2 and 4 percent, respectively. The presented approach classifies 5 types of cervical lesions, the novelty lies in the extensive number of categories, since the literature addresses a more limited number.