DOI: 10.1002/jcu.70313 ISSN: 0091-2751
Noninvasive Preoperative Assessment of Sentinel Lymph Node Metastasis in Breast Cancer Using Super‐Resolution Ultrasound Microvascular Imaging
Ruifang Guo, Xi Liu, Junfeng Zhao, Xiangdong Hu, Linxue Qian ABSTRACT
Purpose
To develop and internally validate a super‐resolution ultrasound microvascular imaging (
SR
‐
UMI
) model for noninvasive preoperative prediction of sentinel lymph node (
SLN
) involvement in breast cancer and to assess incremental value over clinical variables.
Materials and Methods
In this prospective single‐center study (January–June 2025), 162 consecutive patients with pathologically confirmed breast cancer underwent contrast‐enhanced
SR
‐
UMI
of the primary tumor prior to axillary surgery and
SLNB
.
SonoVue
(2.4
mL
) was administered as an antecubital venous bolus. Examinations were performed on an Ultimus
9E
ultrasound system using a 5–15
MHz
linear transducer at low
MI
(0.08). Three predictor sets were evaluated: clinical‐only (9 variables),
SR
‐
UMI
‐only (98 features), and combined (107 predictors). Elastic net–regularized logistic regression (
EN
‐
LR
) was prespecified as the primary model and evaluated using nested stratified cross‐validation (outer 5‐fold, inner 3‐fold) with pooled out‐of‐fold (
OOF
) predictions, assessing discrimination, calibration, and decision‐curve analysis (
DCA
).
Results
SLN
involvement was present in 55/162 patients (34.0%).
EN
‐
LR
achieved
AUC
0.662,
PR
‐
AUC
0.471, and Brier 0.232 for clinical‐only predictors;
SR
‐
UMI
‐only improved performance (
AUC
0.800;
PR
‐
AUC
0.572; Brier 0.185). The combined model performed similarly (
AUC
0.802;
PR
‐
AUC
0.574; Brier 0.185). At the Youden operating point (threshold 0.514), the combined model yielded sensitivity 0.782, specificity 0.766,
PPV
0.632, and
NPV
0.872. Bootstrap
OOF
comparisons showed significant
AUC
improvements for
SR
‐
UMI
‐only versus clinical‐only and combined vs. clinical‐only, whereas combined vs.
SR
‐
UMI
‐only was not significant. Calibration showed a negative intercept and slope > 1 (combined: −0.637, 1.270).
DCA
demonstrated higher net benefit for
EN
‐
LR
than treat‐all/treat‐none across threshold probabilities 0.05–0.60.
Conclusion
SR
‐
UMI
microvascular features enabled good internal‐validation performance and significantly improved discrimination over clinical/biopsy variables. The combined model offered no meaningful gain over
SR
‐
UMI
only. External validation with recalibration is warranted.