Multimodal Gene Expression Deep Learning for Predicting Sentinel Lymph Node Macro-metastasis in Early Breast Cancer: Development and Validation in the SCAN-B Cohort.

Background

This study evaluates DL using GEX and preoperatively available clinical data (PreopClinic) to predict SLNM, and explores their potential for guiding axillary surgery and prognostic assessment.

Methods

We retrospectively included 6,836 clinically node-negative T1-T2 patients with invasive breast cancer who underwent primary surgery from the SCAN-B. Three DL models-a multilayer perceptron, a pathway-informed sparse neural network, and a transformer-were developed using the development set (n=4,625) and evaluated against XGBoost in the independent test set (n=2,211).

Results

The Transformer outperformed other methods for GEX modeling and minimized prior gene selection. In the independent test set, the combined Pre-opClinic+GEX model significantly improved SLNM prediction compared to Pre-opClinic alone (ROC AUC 0.693 vs 0.596, P < 0.001) and identified low-risk patients who might avoid unnecessary SLNB (reduction rate 27.2% at 92.1% sen-sitivity). However, the combined model did not significantly outperform GEX alone. While GEX provided the dominant predictive signal, PreopClinic contributed complementary information with modest numerical gains in clinical utility. Across-subtype training outperformed within-subtype training, particularly in TNBC, where the combined model achieved AUC 0.734 (95% CI: 0.644-0.837). The derived SLNM predictor also provided prognostic information beyond the estab-lished prognostic factors. Although the models were developed primarily using surgical specimen-derived GEX, paired biopsy and surgical-specimen analyses (n=116) demonstrated substantial concordance of transcriptomic patterns and nodal predictions.

Conclusion

These findings highlight the Transformer's robustness against noise and effectiveness in capturing informative transcriptomic features for SLNM pre-diction. The agreement observed between paired biopsy and surgical specimens supports the feasibility of future biopsy-based preoperative applications.