基于动态对比增强 MRI 的影像组学模型预测乳腺癌 TIL（肿瘤浸润淋巴细胞）水平的可行性

Determining tumor-infiltrating lymphocyte (TIL) expression in breast cancer prior to treatment initiation is of considerable clinical significance. MRI demonstrates potential as a valuable adjunct to histopathological assessment. This study investigated the feasibility of developing a radiomics-based predictive model incorporating MRI and clinical features to determine TIL levels in breast cancer.

This multicenter retrospective cohort study enrolled 501 patients across two institutions. A total of 453 patients were utilized for model development, comprising a training cohort (n = 317) and internal validation cohort (n = 136), while 48 patients from an external center constituted the independent test cohort. Radiomics features were extracted and subsequently selected using ANOVA and LASSO regression. Logistic regression algorithms were employed for model construction. Receiver operating characteristic (ROC) curve analysis was performed to evaluate the discriminative performance between low-level and intermediate-high-level TILs, comparing the radiomics model with the integrated clinical-radiomics model.

The MRI-based radiomics model demonstrated robust performance for TIL level prediction across both training and internal validation cohorts, achieving areas under the curve (AUC) of 0.810 (95% confidence interval [CI]: 0.781-0.874) and 0.756 (95% CI: 0.676-0.837), respectively. The integrated clinical-radiomics model exhibited superior discriminative performance with AUCs of 0.828 (95% CI: 0.762-0.858) and 0.824 (95% CI: 0.755-0.893) for the training and validation cohorts, respectively. In the external independent test cohort, the radiomics model and integrated model achieved AUCs of 0.704 (95% CI: 0.558-0.850) and 0.767 (95% CI: 0.633-0.901), respectively.

Radiomics features extracted from dynamic contrast-enhanced MRI, when integrated with clinical characteristics, represent a promising non-invasive approach for predicting TIL levels in breast cancer. This methodology may facilitate clinical decision-making through enhanced pretreatment tumor characterization without requiring invasive tissue sampling.