Acceso usuarios
This study evaluates the application of artificial intelligence (AI) in cervical cancer screening, based on the analysis of 2,383 digitized Pap smear images. These images, sourced from 80 clinical cases in municipalities of Amazonas, Brazil, were meticulously acquired using optical microscopes equipped with digital cameras, and subsequently validated for clarity and detail. Each image underwent a rigorous manual annotation process by researchers under medical supervision, delineating Regions of Interest (ROIs) as 'normal' or 'abnormal' to construct a robust dataset. Two Deep Learning models based on Transformer architecture, namely the Region-based Fully Transformer for End-to-End Object Detection (RF-DETR) and the Real-Time Detection Transformer (RT-DETR), demonstrated promising performance in identifying abnormal cells. The models were thoroughly evaluated using key metrics such as accuracy, sensitivity, specificity, and AUC to assess their diagnostic capabilities. Results demonstrate that AI techniques can successfully support large-scale cervical cancer screening programs, offering valuable assistance in prevention and early diagnosis efforts, and contributing to the reduction of mortality associated with the disease.
