Extracción de caracterı́sticas radiómicas y evaluación de su robustez en imágenes de resonancia magnética de mama

Abstract

[en] This work addresses the evaluation of the robustness of radiomic features (RF) extracted from breast magnetic resonance imaging (MRI). Through the development and implementation of an integrated pipeline, segmentation of regions of interest (ROI), extraction of RF, and assessment of their robustness against variations in segmentation techniques are facilitated. The study examines the reproducibility of different types of RFs, includ- ing shape, first-order statistics, and textural features, against segmentative variations, such as those induced by morphological operations, automatic algorithms, and geometric shapes. The results demonstrate that certain RF, especially those related to shape and first-order statistics, maintain high robustness against simplified segmentative modifications, while textural RF exhibit greater sensitivity to such changes. This analysis provides a basis for informed selection of RFs in machine learning applications and highlights the importance of precision in segmentation for the extraction of reproducible features in radiomic studies. The work also suggests future directions for investigating the influence of the radiologist’s experience on segmentation variability and its impact on RF extraction, offering a promising path for the improvement of predictive models in the diagnosis and treatment of breast cancer.