The use of machine learning to predict cervical cancer progression: an initial approximation

Background: Cervical cancer has two main features: the presence of an oncogenic virus (human papillomavirus, HPV) and the progression of infection to a low-grade squamous intraepithelial lesion (LSIL), which subsequently develops into a high-grade squamous intraepithelial lesion (HSIL) and finally into cervical cancer. In 2014, our laboratory published a study of 75 patients who were diagnosed with LSIL and underwent a second biopsy one year later. In 20 patients, the disease normalized, while in the other 55 patients, LSIL persisted or progressed to HSIL. To measure the prediction of progression, the cells were stained with AgNOR, a proliferation marker that precipitates silver at the sites of nucleolar organizer regions (NORs). In this study, a cutoff value of 3.00 μm2 was set, below which the lesions returned to normal. Materials and Methods: To confirm previous results, the aim of this work is to implement new and modern machine learning algorithms to redefine this cutoff value by applying AgNOR to LSIL. In Python, we ran two models, Support Vector Machine and Logistic Regression, to set a cutoff value for our NOR areas. Results: The cutoff value for each model was 2.82 μm2 for the Support Vector Machine and 2.78 μm2 for Logistic Regression, which are similar to the 3.00 μm2 value reported by our group in 2014. These modern algorithms represent a novel tool that can be used in women with LSIL to predict the future of this lesion. Conclusion: It is possible to predict the progression of LSIL to a persistent lesion or HSIL by staining with the AgNOR technique, measuring the NOR area, and applying machine learning algorithms.