Background: The high false-negative rate (FNR) of sentinel lymph node biopsy (SLNB) alone results in unnecessary axillary lymph node dissection (ALND) in initially clinically node-positive (cN+) breast cancer after neoadjuvant chemotherapy (NAC). This study aimed to reduce the FNR by establishing a longitudinal MRI radiomics-assisted SLNB model compared with SLNB alone. Methods: We established and tested a machine learning algorithm (RMLM) based on pre-NAC, post-NAC and longitudinal MRI radiomics features to detect axillary complete response (ypN0) to neoadjuvant chemotherapy in cN+ breast cancer. In total, 234 eligible patients who underwent SLNB and ALND were retrospectively recruited in the primary cohort (PC), 723 patients were in external validation cohorts (EVC) 1-3, and 81 patients from two multicentre prospective clinical trials (ClinicalTrials.gov, NCT03154749, NCT04858529) were enrolled in the prospective validation cohort (PVC). The U test and least absolute shrinkage and selection operator with tenfold cross-validation were used to select the most significant features. The specificity and FNR of RMLM-assisted SLNB were compared with those of SLNB alone. Results: In the primary (n=234 patients), external validation (n=723) and prospective validation sets (n=81), RMLM achieved excellent performance in detecting metastatic axillary lymph nodes (AUC=0.938 in the primary cohort, 0.842-0.913 in the external cohorts, 0.865 in the prospective validation cohort). RMLM-assisted SLNB achieved an FNR of 4.96% in PC, 5.83-8.08% in EVC1-3, and 9.09% in the prospective validation cohort. The removal of more than two SLNs combined with RMLM could further reduce the FNR to 2.78% in PC, 3.28-7.27% in EVC1-3, and 6.67% in the prospective validation cohort. Conclusions: RMLM combined with SLNB provides a potential tool to evaluate residual nodal disease after NAC and reduce unnecessary ALND.