AI-assisted digital breast tomosynthesis as a decision-support tool for post-neoadjuvant surgical planning in breast cancer.

PURPOSE: Accurate assessment of residual disease after neoadjuvant chemotherapy (NAC) is essential for surgical planning and prevention of incomplete excision. This study evaluated artificial intelligence (AI)-assisted digital breast tomosynthesis (DBT) as a decision-support tool for post-NAC assessment in breast cancer. METHODS: This retrospective study included 151 patients treated with NAC followed by surgery. Post-NAC DBT images were analyzed using Lunit INSIGHT DBT. An AI score threshold of 70 was used to classify patients as having low (< 70) or high (≥ 70) scores. In a paired subgroup of 70 patients, pre- and post-NAC AI scores were compared. Diagnostic performance and univariable and multivariable analyses were performed. RESULTS: The overall pathological complete response (pCR) rate was 41.1% (62/151). Patients with low AI scores had higher pCR rates than those with high scores (49.5% vs. 28.3%, p = 0.010). High AI scores were associated with persistent microcalcifications and residual ductal carcinoma in situ (p < 0.001). Overall accuracy was 58.3% and varied by molecular subtype, with higher performance in triple-negative breast cancer and limited discrimination in HR+/HER2 - tumors. In multivariable analysis, AI score showed borderline significance (p = 0.070). In the paired cohort, AI scores decreased after NAC (p < 0.001), but delta change did not differ between pCR and non-pCR groups (p = 0.877). CONCLUSION: AI-assisted DBT may have potential complementary value as a decision-support tool for post-NAC surgical planning, particularly in triple-negative breast cancer and in cases with residual microcalcifications.