Beyond Masses and Calcifications: A Review of Architectural Distortion Detection for Early Breast Cancer Diagnosis | ||||
Benha Journal of Engineering Science and Technology | ||||
Volume 2, Issue 1, July 2025, Page 126-136 PDF (401.58 K) | ||||
Document Type: Research Article | ||||
DOI: 10.21608/bjest.2025.445862 | ||||
![]() | ||||
Authors | ||||
Sameh E. Ibrahim* 1; Wael A. Mohamed1; Ahmed F. Elnokrashy1, 2 | ||||
1Department of Electrical Engineering, Benha Faculty of Engineering, Benha University, 13511, Benha, Egypt | ||||
2School of Information Technology and Computer Science, Nile University, 26th of July Corridor, Sheikh Zayed City | ||||
Abstract | ||||
Architectural distortion (AD) in mammograms is a subtle yet critical marker of early breast cancer, often appear up to two years before other findings like masses and calcifications. Despite its clinical significance accounting for up to 45% of missed breast cancers AD remains diagnostically challenging due to its subtle patterns, overlap with dense breast tissue, and the reliance on radiologist expertise. This review synthesizes the evolution of AD detection methodologies, from early handcrafted feature-based approaches to advanced machine learning (ML) and deep learning (DL) techniques. The performance of these methods is evaluated, highlighting the superior results achieved by modern deep learning models, such as U-Net and attention-based networks, which automate feature extraction. However, challenges persist, including limited annotated datasets and high false-positive rates, which hinder clinical adoption. The need for standardized datasets, multimodal imaging integration, and collaborative efforts to develop AD-specific datasets and hybrid AI-human workflows is emphasized. Bridging technical innovations with clinical practice is essential to improving early breast cancer diagnosis and ensuring more accurate and widespread detection of architectural distortion. | ||||
Keywords | ||||
Architectural distortion; Breast cancer; Machine learning; Deep learning | ||||
Statistics Article View: 47 PDF Download: 30 |
||||