P11.24.a Deep Learning Radiomic Model Predicts Intratumoral Tertiary Lymphoid Structures in Breast Cancer Brain Metastasis

Abstract BACKGROUND Intratumoral tertiary lymphoid structures (iTLS) in breast cancer brain metastases (BCBM) are associated with improved survival and may influence treatment decision-making. However, their non-invasive detection remains challenging in BCBM. We aim to develop a non-invasive model using baseline 3d contrast-enhanced MRI to predict the iTLS status and the prognosis. MATERIAL AND METHODS A total of 120 patients with BCBM who underwent surgery were retrospectively recruited from our centre between Decermber 2015 and April 2024 and divided into training and validation sets. TLSs were evaluated, differences in progression-free survival (PFS) and overall survival (OS) between groups were calculated using the Kaplan-Meier method. Immunohistochemistry and multiplex immunofluorescence (mIF) were used to assess TLSs heterogeneity. After feature dimensionality and selection, corresponding features were used to construct deep learning radiomic (DLR) models for diagnosing iTLS based on preoperative MR. The performances of models were assessed using the area under the receiver operating characteristic curve (AUC). The log-rank test was used to evaluate the prognostic value of the DLR model. RESULTS The presence of iTLS was identified in 46.7% (n=120) patients. TLS was positively associated with OS (p=0.0172) and PFS (p=0.0161) in the human epidermal growth factor receptor type 2-positive subtype, and with prolonged OS (p=0.0482) in the triple-negative breast cancer subtype. The DLR model demonstrated excellent performance in predicting the presence of iTLS in training (AUC=0.92, 95% CI: 0.88, 0.95) and validation set (AUC=0.86, 95% CI: 0.82, 0.91). The DLR model-predicted iTLS group has favourable overall survival (HR=0.67; 95% CI: 0.47, 0.91; p=0.01) and progression-free survival (HR=0.65; 95% CI: 0.49, 0.87; p=0.009) in the validation set. CONCLUSION The DLR model has indicated accurate prediction of iTLS status, which may assist in the risk stratification and immunotherapy candidate selection for patients with BCBM in clinical practice.