A total of 240 patients with node-positive breast cancer (mean age, 46.36 ± 10.23 years; range, 25–70 years) were included (Table 1). Among them, 129 (53.75%) cases achieved axillary pCR and 111 (46.25%) cases had residual metastasis in axillary LNs. Furthermore, 33 (13.75%) cases were found to have positive LN(s) in axilla level III while 207 (86.25%) cases had negative axilla level III after NAT. Patients with positive axilla level III after NAT were significantly more likely to have higher clinical tumor stage and nodal stage at initial, as well as negative HER2 expression. Table 1 presents the baseline clinicopathologic characteristics of the patients.

According to pathologic result of ALND, among 111 patients with axillary non-pCR, 47 (42.34%) cases had residual metastasis only in axilla level I; 31 (27.93%) cases had positive LNs in axilla level II and no positive LNs in axilla level III; positive LNs were found in axilla level III in 33 (29.73%) cases. Regarding axilla US diagnosis, no suspicious LNs were found in 120 cases, while 120 cases had positive LNs on US. Among the 120 cases with positive LNs on US, 73 cases were found to have positive LNs only in axilla level I, 36 cases in axilla level II, and 11 cases in axilla level III. The kappa value of axilla US diagnosis and ALND result for determining axilla level with positive LNs is 0.39 (p < 0.001), indicating a fair agreement (Table 2). According to ALND results, axilla US accurately identified 75.19% (97/129) of cases with axillary pCR, 61.70% (29/47) of cases with residual metastasis only in axilla level I, 35.48% (11/31) of cases in axilla level II, and 24.24% (8/33) of cases in axilla level III.

In this study, the negative/positive classification of axilla level III according to the ALND result served as the gold standard, while the axilla level with positive LNs diagnosed by US was considered as the independent variable. Based on ROC curve analysis, the performance of US diagnosis for LNs in axilla level III was found to be limited, with an area under ROC curve (AUC) of 0.67, an accuracy of 67.57%, a sensitivity of 51.52%, and a specificity of 74.36%. The cutoff value was determined to be a positive axilla level on US of > I, indicating that suspicious LNs in axilla level II or III on US were considered as predictors of pathologically positive axilla level III.

For cases with negative axilla level III after NAT, the SWVmax, SWVmin, and SWVmean values of breast lesions were 3.18 ± 1.62 m/s, 2.21 ± 0.70 m/s, and 2.70 ± 1.11 m/s, respectively. In contrast, for breast lesions with positive axilla level III, the SWVmax, SWVmin, and SWVmean values were 5.97 ± 1.90 m/s, 2.89 ± 0.85 m/s, and 4.43 ± 1.27 m/s, respectively. All of these values were significantly higher than those for cases with negative axilla level III (p < 0.001), as indicated in Table 3 and Fig. 2. According to the ROC curve analysis, SWVmax exhibited the best performance in identifying positive axilla level III after NAT, with an AUC of 0.79. The optimal cutoff value of SWVmax was 4.21 m/s, with an accuracy of 77.48%, a sensitivity of 84.85%, and a specificity of 74.36% (Table 4). SWVmax demonstrated significantly superior performance for determining nodal status in axilla level III compared to SWVmin and SWVmean (SWVmax vs. SWVmin, p < 0.001 and SWVmax vs. SWVmean, p = 0.016, Delong’s test), as shown in Fig. 3.

Clustered box plot of SWE characteristics for both negative and positive axilla level III after NAT. Abbreviations: SWE, shear wave elastography; NAT, neoadjuvant therapy; SWV, shear wave velocity; LN, lymph node

ROC curves for using SWE in diagnosing LNs in axilla level III after NAT in patients with node-positive breast cancer. Abbreviations: ROC, receiver operating characteristic; SWE, shear wave elastography; LN, lymph node; NAT, neoadjuvant therapy; SWV, shear wave velocity

In our study, axilla level with suspicious LNs on US and breast SWVmax were identified as associated variables with the pathologic status of axilla level III after NAT. A higher axilla level with positive LNs on US (OR, 1.36, 1.75, and 26.06 for positive axilla level I, II, and III on US, respectively) and a higher SWVmax of the breast lesions (OR, 1.58) were found to be independently associated with pathologically positive axilla level III. Consequently, a combined US model was developed based on axilla US diagnosis and breast SWE characteristics, using multivariate LR analysis. The combination of axilla level with suspicious LNs on US and the SWVmax of the breast lesion resulted in significantly higher performance than conventional axilla US alone (AUC, 0.86 vs 0.69, p < 0.001, Delong’s test) for identifying positive LNs in axilla level III after NAT, with an enhanced accuracy of 81.08%, a remarkably improved sensitivity of 93.94%, and a comparable specificity of 75.64%, at a predictive probability cutoff value of 0.18 (Table 5 and Fig. 4).

ROC curves for using SWE in determining the status of axilla level III after NAT in patients with node-positive breast cancer. Abbreviations: ROC, receiver operating characteristic; SWE, shear wave elastography; NAT, neoadjuvant therapy; US, ultrasound

The p-value obtained using the Hosmer–Lemeshow goodness-of-fit test was 0.085 (X2 = 13.90), indicating a good fit of the combined model. Calibration curve analysis demonstrated good agreement between the observed and predictive status of LNs in axilla level III after NAT in breast cancer patients with positive LNs at the initial diagnosis (slope, 1.12), as depicted in Fig. 5. Additionally, the decision curve analysis revealed that, when the probability threshold was 0.00–0.57 and 0.75–0.96, clinical decision-making based on the combined model offered superior overall benefits compared to the all-or-none strategy, as depicted in Fig. 6. Figures 7 and 8 illustrate the effectiveness of this combined model.

Calibration curve for the predictive model

Decision curve for the clinical benefit of the predictive model

The characteristics of breast SWE and axilla US in a cases with positive axilla level III after NAT. A 53-year-old case with T2N3 Luminal-A breast cancer was found to have residual metastasis in 14 LNs in axilla level I, 3 in axilla level II, and 2 in axilla level III according to ALND result. A A solid, hypo-echogenic breast lesion measuring 18.1 × 4.2 mm was found on US. B The breast lesion showed a SWVmax value of 5.30 m/s on the SWE-velocity map. C–D Several suspicious small LNs were identified in axilla level I by using axilla US, characterized by ratio of long axis diameter to short axis diameter < 2, irregular shape, ill-defined margin, effacement of the fatty hilum, and absence of hilar blood flow. No LNs were detected in axilla level II or III on US. Based on the model combing breast SWVmax of 5.30 m/s and positive axilla level I on US, the predictive result indicated the presence of residual metastasis in axilla level III. SWE, shear wave elastography; US, ultrasound; NAT, neoadjuvant therapy; LN, lymph node; ALND, axillary lymph node dissection; SWV, shear wave velocity

The characteristics of breast SWE and axilla US in a cases with negative axilla level III after NAT. Following NAT, a 50-year-old patient with T2N2 Luminal-B2 breast cancer underwent ALND, during which 21 lymph nodes were removed from axilla levels I-III. According to ALND result, only 1 LNs in axilla level II was found to have residual metastasis. A A solid, hypo-echogenic breast lesion with interior calcification measuring 8.6 × 7.4 mm was found on greyscale US. B The breast lesion showed a SWVmax value of 3.17 m/s on the SWE-velocity map. C–D Several LNs without suspicious features were identified in axilla level I, and no LNs were detected in axilla level II or III on US. Based on the model combing breast SWVmax of 3.17 m/s and negative axilla on US, the predictive result indicated the absence of residual metastasis in axilla level III. SWE, shear wave elastography; US, ultrasound; NAT, neoadjuvant therapy; ALND, axillary lymph node dissection; LN, lymph node; SWV, shear wave velocity