Akademik Veri Yönetim Sistemi
Engineering Sciences, cilt.20, sa.4, ss.129-147, 2025 (Hakemli Dergi)
Breast cancer is the commonly diagnosed cancer in women all overthe world, and its prevalence is constantly increasing despitesignificant advancements in the area of early diagnosis and individualtreatment approaches. Nevertheless, present-day workflows in diagnosticinterventions are struggling with problems such as overdiagnosis inpopulations with low risks, growing workloads among radiologists andpathologists, and inconsistencies in the interpretation of the findingsof the imaging and pathological studies. In that regard, artificialintelligence (AI) has proven to be an effective solution to thesedrawbacks by enhancing image analysis, automating the working processesthat consume a lot of labor, and facilitating clinical decision-making.This paper provides a narrative review of the recent AI implementationin breast cancer screening and diagnosis, including malignancy detectionand classification, tumor segmentation, prediction of molecular subtype,and recurrence or metastatic risk. The data sources are analyzed bothin imaging and non-imaging, which are mammography, ultrasound, magneticresonance imaging (MRI), histopathology, clinical variables, and multi-modal data integration. Also, the reviewed articles identify explainableartificial intelligence (XAI) methods, including SHAP, Grad-CAM, andLIME, as central to improving the transparency, interpretability, andconfidence clinicians have in AI-assisted systems. On the whole, thecurrent evidence indicates that AI-based tools have the potential toincrease the level of diagnostic accuracy, minimize inter-observervariability, and provide a personalized risk evaluation and treatmentplanning. However, there are still multiple obstacles to widespreadclinical implementation such as heterogeneity of datasets, a lack ofexternal and prospective validation, interpretability issues, andconstraints based on real-world application. Future studies must,therefore, focus on the creation of more and better-quality data,standard assessment guidelines, solid explainability models, and futureclinical trials to allow the safe, productive, and fair integration ofAI into regular breast cancer care.