Study of radiomics and artificial intelligence applications in multimodal breast imaging: from conventional to advanced imaging

Abstract

Introduction: Breast cancer remains the leading cause of cancer-related mortality among women. While radiological screening is vital, current diagnostic pathways face limitations due to the subjective nature of image interpretation and the invasiveness of confirmatory biopsies (CNB/VABB), leading to high costs and patient distress. Objectives: This thesis explores the application of Artificial Intelligence (AI) and Radiomics across four key breast imaging modalities: Mammography (MX), Ultrasound (US), MRI, and Contrast-Enhanced Mammography (CEM). The primary aim is to enhance the differentiation between benign and malignant lesions, specifically focusing on reducing unnecessary biopsies for BI-RADS R4a microcalcifications and B3 lesions (ADH), and validating CAD systems for clinical support. Materials and Methods: A standardized radiomics workflow was applied to integrated retrospective and prospective datasets. The process included image acquisition, Region of Interest (ROI) segmentation, and the extraction of quantitative features (morphological, textural, and high-order). Predictive models were built using Machine Learning techniques (e.g., LASSO regression) and validated against histopathological gold standards. Results and Conclusions: The synergy between clinical data and radiomic features demonstrates significant potential in improving diagnostic accuracy beyond qualitative assessment. Radiomics represents a promising frontier in precision medicine, providing non-invasive biomarkers to personalize patient management, minimize recall rates, and reduce overtreatment in high-risk but potentially benign lesions.

Ähnliche Arbeiten