Artificial intelligence in the radiologic assessment of ductal carcinoma <i>in situ</i>: a systematic review.
Authors
Affiliations (2)
Affiliations (2)
- University of the Incarnate Word School of Osteopathic Medicine, San Antonio, TX, United States.
- Baptist University College of Osteopathic Medicine, Memphis, TN, United States.
Abstract
Ductal carcinoma <i>in situ</i> (DCIS) accounts for approximately 25% of new breast cancer diagnoses and carries a 20% to 50% risk of post-surgical upstaging to invasive cancer, complicating risk stratification and raising concerns about overtreatment. Mammography, the primary imaging modality, is limited by low sensitivity in dense breasts and low-grade lesions. Supplemental imaging (ultrasound, magnetic resonance imaging) can improve detection but is often costly and less accessible. Artificial intelligence (AI) has emerged as a promising approach to enhance DCIS radiologic assessment. A systematic search of several databases was conducted to determine the role of AI technologies in supporting radiological assessment of DCIS following PRISMA guidelines. Of 311 studies identified, 46 met inclusion criteria. AI shows substantial promise in improving detection, classification, preoperative risk stratification, and molecular inference, with area under the curves (AUCs) ranging from 0.70 to 0.97, sensitivities of 80-96%, and specificities up to 93%. Overall, temporal, multiphase, and spatially aware models outperformed conventional 2D approaches. The findings underscore the potential for clinical integration, inform best practices, and identify critical gaps to guide future studies and development of standardized, validated AI tools for DCIS management. Limitations included retrospective design, small cohort of DCIS-specific datasets, minimal external validation, generalizability concerns, and weaker performance relative to invasive breast cancer.