Development of a preoperative prediction model based on spectral CT to evaluate axillary nodal burden in patients with early-stage breast cancer

Background: Axillary lymph node (ALN) status is an important factor affecting the surgery planning. We developed a new prediction model based on spectral CT to evaluate the axillary nodal burden in patients with early-stage breast cancer. Methods: A total of 146 early-stage breast cancer patients with pathologically confirmed axillary lymph node (ALN) status who underwent spectral CT scans were retrospectively enrolled and stratified sampling to categorize into a training cohort (n = 102) and validation cohort (n = 44). These patients were classified into low (stage pN0 or pN0[i+] or pN1mi) and high (stage pN1–3) axial nodal burden subgroups. The morphologic criteria and quantitative spectral CT parameters of the most suspicious lymph node were measured and compared between high axillary lymph node (HANB) and low axillary lymph node (LANB). Univariate analyses were performed to identify the association between factors and HANB. Least absolute shrinkage and selection operator (Lasso) was used to further screen predictive indicators to build a logistic model. The receiver operating characteristic (ROC) curve and decision curve analysis (DCA) were used to evaluate the models. Results: Univariate analysis demonstrated that 16 risk factors were associated with HANB (P < 0.05). For arterial phase, the LASSO regression analysis identified the 5 most powerful factors (the longest-to-shortest diameter ratio, fatty hilum, water concentration, nZ eff and λ HU ), and 4 most powerful factors (the shortest diameter, fatty hilum, nIC and nZ eff ) for venous phase. When combining arterial phase and venous phase, four variables (the fatty hilum, arterial phase nZ eff , venous phase nZ eff and venous phase nIC) were included. In the validation cohort, the AUCs for arterial phase spectral CT model, venous phase spectral CT model and combined arterial-venous phase spectral CT model were 0.89, 0.91, 0.91, respectively. There was no significant difference in AUCs among the three models (DeLong test, P > 0.05 for each comparison). Conclusion: We developed a Lasso-logistic model that combined morphologic and quantitative spectral CT parameters of the ALN for preoperative prediction of ALN status in patients with clinical T1/2N0 breast cancer, presenting a non-invasive and highly useful predictive tool for individual preoperative prediction of ALN status in patients with early-stage breast cancer.