Abstract:
Background: Predicting pathological complete response (pCR) to neoadjuvant chemotherapy (NAC) in breast cancer is essential for personalizing treatment. AI-assisted ultrasound has emerged as a promising non invasive approach, yet a systematic review specifically evaluating the diagnostic accuracy of AI-assisted ultrasound for pCR prediction remains limited.
Methods: We conducted a systematic review and meta-analysis of studies evaluating AI based models including deep learning (DL), machine learning (ML), and radiomics—applied to ultrasound imaging for pCR prediction in breast cancer patients undergoing NAC. Four databases (PubMed, Scopus, Web of Science, Cochrane) were searched from inception to May 2025. Study quality was assessed using QUADAS-2. A bivariate random-effects model was used to pool sensitivity, specificity, diagnostic odds ratio (DOR), and the area under the summary receiver operating characteristic (SROC) curve.
Results: Twenty studies were included in the qualitative synthesis, of which nine were included in meta-analysis (n = 2,271 patients). The pooled sensitivity was 0.82 (95% CI: 0.73–0.88) and specificity 0.85 (95% CI: 0.80–0.89), with an SROC AUC of 0.90 (95% CI: 0.87–0.93) and DOR of 26 (95% CI: 13–54). The deep learning subgroup yielded an AUC of 0.86. A negative test result reduced post-test probability of pCR to approximately 11%, underscoring strong performance. Significant heterogeneity was observed (I² = 86%), partly due to threshold effects. No publication bias was detected (Deeks' test p = 0.88).
Conclusion: AI-assisted ultrasound demonstrates high diagnostic accuracy for predicting pCR in breast cancer, offering a practical, radiation-free complement to conventional imaging. Prospective multicenter validation and standardization of protocols are needed before routine clinical implementation.
