To evaluate current evidence on AI’s application in DCIS imaging, focusing on diagnostic performance and limitations.
Approach:
Systematic Review: A systematic search of MEDLINE Ovid, PubMed, and Web of Science was conducted following PRISMA guidelines to identify studies on AI in DCIS radiologic assessment, resulting in 46 studies meeting inclusion criteria.
Key Findings:
AI shows promise in improving detection, classification, preoperative risk stratification, and molecular inference for DCIS.
Area under the curves (AUCs) for AI models ranged from 0.70 to 0.97, with sensitivities of 80–96% and specificities up to 93%.
Temporal, multiphase, and spatially aware models outperformed conventional 2D approaches.
Interpretation:
The findings suggest potential for clinical integration of AI in DCIS management, while also highlighting critical gaps for future studies.
Limitations:
Retrospective design of included studies.
Small cohort of DCIS-specific datasets.
Minimal external validation.
Concerns regarding generalizability.
Weaker performance of AI models relative to invasive breast cancer.
Conclusion:
AI has the potential to enhance various aspects of DCIS diagnosis and management, but further validation and standardization are needed.
