Processamento digital de imagens aplicado à mamografia: uma abordagem baseada em técnicas de segmentação, realce e marcação de lesões

Abstract

This study investigates the application of a digital image processing pipeline to optimize mammogram analysis, using the public INbreast dataset and a private clinical collection. The methodology comprises sequential steps, including automatic breast region cropping, intensity normalization via percentile truncation, and adaptive contrast enhancement using the CLAHE algorithm. Subsequently, morphological and contour-based criteria are applied to generate bounding boxes over regions of interest. The results show that while preprocessing techniques significantly improve the visual quality of images for medical interpretation or as input for computer-aided diagnosis (CAD) systems, the automatic detection stage exhibits limitations in cases of low contrast or lesions with complex morphology. The work concludes that the classical approach serves as a fundamental basis, but its robustness is limited, highlighting the need for integration with machine learning models for more accurate and generalizable detection.