Analysis of the Correlation and Prognostic Significance of Tertiary Lymphoid Structures in Breast Cancer: A Radiomics‐Clinical Integration Approach
Kezhen Li, Juan Ji, Simin Li, Man Yang, Yurou Che, Zhu Xu, Yiyao Zhang, Mei Wang, Zengyi Fang, Liping Luo, Chuan Wu, Xin Lai, Juan Dong, Xinlan Zhang, Na Zhao, Yang Liu, Weidong Wang- Radiology, Nuclear Medicine and imaging
Background
Tertiary lymphoid structures (TLSs) are potential prognostic indicators. Radiomics may help reduce unnecessary invasive operations.
Purpose
To analyze the association between TLSs and prognosis, and to establish a nomogram model to evaluate the expression of TLSs in breast cancer (BC) patients.
Study Type
Retrospective.
Population
Two hundred forty‐two patients with localized primary BC (confirmed by surgery) were divided into BC + TLS group (N = 122) and BC − TLS group (N = 120).
Field Strength/Sequence
3.0T; Caipirinha‐Dixon‐TWIST‐volume interpolated breath‐hold sequence for dynamic contrast‐enhanced (DCE) MRI and inversion‐recovery turbo spin echo sequence for T2‐weighted imaging (T2WI).
Assessment
Three models for differentiating BC + TLS and BC − TLS were developed: 1) a clinical model, 2) a radiomics signature model, and 3) a combined clinical and radiomics (nomogram) model. The overall survival (OS), distant metastasis‐free survival (DMFS), and disease‐free survival (DFS) were compared to evaluate the prognostic value of TLSs.
Statistical Tests
LASSO algorithm and ANOVA were used to select highly correlated features. Clinical relevant variables were identified by multivariable logistic regression. Model performance was evaluated by the area under the receiver operating characteristic (ROC) curve (AUC), and through decision curve analysis (DCA). The Kaplan–Meier method was used to calculate the survival rate.
Results
The radiomics signature model (training: AUC 0.766; test: AUC 0.749) and the nomogram model (training: AUC 0.820; test: AUC 0.749) showed better validation performance than the clinical model. DCA showed that the nomogram model had a higher net benefit than the other models. The median follow‐up time was 52 months. While there was no significant difference in 3‐year OS (P = 0.22) between BC + TLS and BC − TLS patients, there were significant differences in 3‐year DFS and 3‐year DMFS between the two groups.
Data Conclusion
The nomogram model performs well in distinguishing the presence or absence of TLS. BC + TLS patients had higher long‐term disease control rates and better prognoses than those without TLS.
Evidence Level
2
Technical Efficacy
Stage 2