Title : Citobot: An AI-driven solution for cervical cancer screening and public health innovation in Latin America
Abstract:
Cervical cancer remains one of the main causes of illness and death among women worldwide, particularly in low- and middle-income countries, despite being entirely preventable through timely screening and vaccination. Delays in diagnostic result delivery and low adherence to cytology-based programs contribute to late detection and poor outcomes. Artificial intelligence, specifically deep learning, has emerged as a promising approach to improve early detection, reduce diagnostic variability, and expand access in resource-limited settings.
This study aimed to validate the diagnostic performance of the CITOBOT artificial intelligence system, a model designed for automated cervical image classification and risk stratification, as a support tool for cervical cancer screening in Colombia. A cross-sectional study was conducted with 650 women screened at a public hospital in Cali between February 2023 and July 2025. Colposcopy and biopsy results were used as the diagnostic gold standard. The dataset included 2,648 cervical images labeled as No Risk and At Risk. The model was developed using the InceptionV3 architecture with transfer learning, data augmentation, and five-fold cross-validation. Preprocessing included normalization, segmentation with the Segment Anything Model, and class balancing.
CITOBOT achieved an accuracy of 94.3%, sensitivity of 93.4%, and specificity of 94.9%, with an area under the receiver operating characteristic curve of 0.98. The positive predictive value was 92.9% and the negative predictive value 96.2%. Precision and recall were well balanced, and misclassification was minimal. The model demonstrated stable convergence without overfitting, confirming its robustness and generalizability.
CITOBOT showed high diagnostic accuracy and balanced predictive performance for cervical cancer risk stratification. Its integration into low-resource clinical settings could reduce diagnostic delays, strengthen early detection, and help decrease cervical cancer–related inequities in Latin America.
