Avanços em modelos computacionais para predição da sobrevida para o câncer de mama feminino

Abstract

Breast Cancer is one of the leading causes of death among women, often diagnosed and treated at advanced stages, which contributes to high mortality rates. The application of Machine Learning methods has shown great potential in predicting outcomes in oncological diseases. This study aimed to identify and evaluate prognostic attributes for predicting the survival of female BC patients through computational models. The research involved the construction and analysis of a clinical Database containing information from patients in the Zona da Mata region of Minas Gerais. After data collection, preprocessing steps were carried out, beginning with a simple imputation phase followed by a second phase of inferring missing values, considering descriptive analysis, data correlation, and the clinical significance of each variable. Variables with a correlation above 70% were reanalyzed, taking into account their clinical relevance, and the most significant variables were selected. These variables were then subjected to feature selection methods, which were essential to enhance model accuracy and identify critical factors in survival prediction. Based on this database, different ML methods were evaluated, including linear models, such as Cox Proportional-Hazards Regression, both non-penalized and penalized (Lasso and Elastic Net), as well as Survival Support Vector Machine. Non-linear models, such as Random Survival Forest, Gradient Boosting Survival, and Kernel Survival Support Vector Machine, were also tested. Performance evaluation, using specific survival analysis metrics, showed that Random Survival Forest outperformed the other models. The study also highlighted the importance of data preprocessing and the application of feature selection methods, which were crucial in identifying relevant prognostic variables for clinical practice. Additionally, the impacts of this computational approach on clinical decision-making were discussed. The results underscore the potential of computational models in predicting the survival of female BC patients, and this research contributes to the advancement of computational oncology, with the potential to improve prognoses and patients’ quality of life.