Risk of breast cancer in the mexican population: a radiomics approach

Abstract

Breast cancer is a significant global health concern, especially among women, with rising incidence rates in specific populations. Low screening rates contribute to this alarming trend, emphasizing the need to improve breast cancer risk prediction and enhance screening outcomes. This thesis explores the potential of image-based models and machine learning techniques to address limitations in traditional risk assessment models and leverage the rich information available in mammography images. A larger dataset, including diverse cases with breast cancer diagnoses, is recommended to improve accuracy and generalizability. In addition, extracting additional image-based features to characterize breast anatomy could provide valuable insights. The outcomes of this research can contribute to personalized medicine approaches and improve breast cancer risk prediction, leading to early detection, timely interventions, and improved patient outcomes. This study showed successful segmentation and extraction of 78 features per image (first and second order), and the methodology's performance with a machine learning Cox model achieved an AUC of 0.76. Furthermore, the Kaplan-Meier curve significantly differed between the low-risk and high-risk groups. The advantage of using a Cox model is its ability to identify the most discriminative features, which in this case were three features associated with the physiological characteristics of the patients. This thesis provides a roadmap for further investigation, emphasizing the importance of larger datasets, technique refinement, and exploration of population-specific characteristics to develop more effective breast cancer screening and prevention strategies.