Predicting axillary residual disease after neoadjuvant therapy in breast cancer using baseline MRI and ultrasound

To predict axillary node residual disease in women treated for node-positive breast cancer (BC) by neoadjuvant therapy (NAT), using breast BI-RADS MRI features and axillary ultrasound at baseline.

In this single-center, retrospective study, women with node-positive BC who underwent NAT between 2016 and 2021 were included. Pre-treatment axillary US and breast MRIs were evaluated using the BI-RADS lexicon and T2 features, including Breast Edema Score. Univariate and multivariate logistic regression analyses were conducted for the prediction of axillary residual disease (ARD). A multivariable model based on logistic regression was trained and evaluated on randomly split train and test sets (7:3 ratio).

Out of the 141 women, 41% had post-NAT ARD. Axillary metastasis was independently associated with luminal subtype (odds ratio (OR), 25.5; p < 0.001), anterior tumor location (OR, 14.1; p = 0.008), and cortical thickening ≥ 7 mm (OR, 6.09; p = 0.002). Intratumoral T2 high signal intensity was protective (OR, 0.16; p = 0.006), while Ki67 had a marginal association (p = 0.064). In the training and test sets, the model, which is available online, achieved AUCs of 0.860 (95% CI: 0.783-0.936) and 0.843 (95% CI: 0.714-0.971), respectively. Anterior depth location and cortical thickening greater than 7 mm were also independently associated with post-NAT axillary burden.

Adjusting for BC subtype and KI-67 index, the anterior third location of BC, a cortical thickness greater than 7 mm, and the absence of intratumoral T2 hyperintensity is predictive of ARD after NAT.

Question What baseline imaging-based predictive models can identify patients at risk of persistent nodal disease after neoadjuvant therapy? Findings Baseline US cortical thickness superior to 7 mm, anterior tumor location, and absence of an intratumoral high signal on T2-weighted MRI predict residual axillary disease. Clinical relevance Our predictive model, available online at: litoic.shinyapps.io/LNPred_Apps , including breast cancer subtype, Ki-67 index level, breast cancer location, intratumoral signal intensity on T2WI, and initial lymph node thickness, could guide post-NAT axillary management.