DIAGNOSTIC ACCURACY OF ARTIFICIAL INTELLIGENCE IN THE UPSTAGING PREDICTION OF DUCTAL CARCINOMA IN SITU: A SYSTEMATIC REVIEW AND META-ANALYSIS

Background: Ductal carcinoma in situ (DCIS) may be upstaged to invasive carcinoma after surgical excision. Predicting upstaging before surgery may help guide treatment planning. Artificial intelligence (AI)-based imaging has been investigated for this purpose, but its overall performance remains unclear.
Objective: To evaluate the diagnostic accuracy of AI-based systems for predicting DCIS upstaging to invasive carcinoma.
Materials and Methods: A systematic review and meta-analysis was conducted following PRISMA 2020 and registered in PROSPERO (CRD420261516994). PubMed, Scopus, EBSCOhost, ProQuest, and Web of Science were searched from inception to September 2026. Studies evaluating AI-based imaging models for DCIS upstaging were included. Pooled sensitivity and specificity were calculated using a bivariate random-effects model. Risk of bias was assessed using QUADAS-2.
Results: Six studies involving 701 participants were included. Pooled sensitivity was 0.77 (95% CI, 0.57–0.89) and specificity was 0.78 (95% CI, 0.64–0.87), with substantial heterogeneity (I² = 84.0% and 90.6%, respectively). The pooled log diagnostic odds ratio (DOR) was 2.15 (95% CI, 1.64–2.65). The SROC analysis showed moderate overall discriminative performance. No significant differences were observed across imaging modalities, while AI model type significantly affected specificity but not sensitivity. Deeks’ funnel plot suggested possible publication bias (p = 0.046).
Conclusion: AI-based imaging showed moderate accuracy for predicting DCIS upstaging. However, substantial heterogeneity and methodological limitations limit confidence in these findings. Further prospective studies with external validation are needed.