MRI影像组学联合基因标签预测乳腺癌新辅助化疗后病理完全缓解
本研究回顾性纳入112例乳腺癌患者,提取新辅助化疗前MRI的100个影像组学特征,联合免疫相关23基因标签(IRSN-23)构建支持向量分类模型预测病理完全缓解。训练队列中联合模型AUC为0.93(95%CI 0.86-1.00),优于单独IRSN-23的0.83和单独MRI影像组学的0.76;验证队列中联合模型AUC为0.89(95%CI 0.69-1.00)。联合模型可能辅助治疗决策,但生存关联为探索性。
要点
- 新辅助化疗前MRI影像组学联合IRSN-23基因标签可提高乳腺癌pCR预测效能。
- 联合模型在训练队列AUC达0.93,验证队列AUC为0.89,均高于单一模态。
- 模型预测pCR与无复发生存和总生存的关联未达统计学显著,仅为探索性。
核心结论
- MRI影像组学联合IRSN-23基因标签较任一单独模态提高pCR预测效能,可能支持治疗规划,但生存关联应视为探索性。
研究设计与样本
- 回顾性研究纳入112例经组织学确诊的乳腺癌女性(pCR 21例,非pCR 91例),均接受新辅助化疗前MRI并有肿瘤组织来源的IRSN-23评分。
- 患者分为训练队列100例和验证队列12例,从新辅助化疗前MRI提取100个影像组学特征。
- 采用支持向量分类器训练三种模型:单独影像组学、单独IRSN-23、联合影像基因组模型,以ROC分析和AUC评估。
诊断标准与关键数值
- 训练队列AUC(95%CI):IRSN-23为0.83(0.73-0.92),MRI影像组学为0.76(0.63-0.89),联合模型为0.93(0.86-1.00)。
- 验证队列AUC(95%CI):IRSN-23为0.78(0.49-1.00),MRI影像组学为0.59(0.20-0.99),联合模型为0.89(0.69-1.00)。
- 训练队列中联合模型显著优于MRI影像组学(p=0.007)和IRSN-23(p=0.031)。
临床应用与阅片要点
- 新辅助化疗前MRI影像组学特征联合IRSN-23基因标签可辅助预测乳腺癌pCR,为治疗规划提供参考。
- 模型预测pCR与无复发生存和总生存的关联未达统计学显著,不能据此判断预后。
局限性
- 回顾性研究设计,验证队列仅12例,样本量小且分组不均衡。
- 生存分析中模型预测pCR与RFS、OS的差异未达统计学显著,作者声明生存关联为探索性。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
英文原文(摘要)· 点击展开
PURPOSE: Pathological complete response (pCR) after neoadjuvant chemotherapy (NAC) is strongly associated with improved survival. Radiogenomic approaches integrating MRI. radiomics with gene signatures may enhance pretreatment prediction of therapeutic.
RESPONSE: This study aimed to develop a radiogenomic model for predicting pCR in breast cancer using MRI radiomics combined with an immune-related 23-gene signature (IRSN-23).
MATERIALS AND METHODS: We retrospectively analyzed 112 women with histologically confirmed breast cancer (pCR, n = 21; non-pCR, n = 91) who underwent pre-NAC MRI and had IRSN-23 scores derived from tumor tissue. Patients were assigned to training (n = 100) and validation (n = 12) cohorts. From pre-NAC MRI, 100 radiomic features were extracted. A support vector classifier was trained to predict pCR using three models: radiomics alone, IRSN-23 alone, and the combined radiogenomic model. Receiver operating characteristic (ROC) analysis and areas under the curve (AUC) were calculated. Kaplan-Meier survival analyses assessed relapse-free survival (RFS) and overall survival (OS) based on model-predicted pCR.
RESULTS: In the training cohort, AUCs (95 % confidence interval [CI]) were 0.83 (0.73-0.92) for IRSN-23, 0.76 (0.63-0.89) for MRI radiomics, and 0.93 (0.86-1.00) for the combined model. In the validation cohort, corresponding AUCs were 0.78 (0.49-1.00), 0.59 (0.20-0.99), and 0.89 (0.69-1.00). The combined model significantly outperformed MRI (p = 0.007) and IRSN-23 (p = 0.031) in the training cohort. Differences in RFS and OS according to model-predicted pCR did not reach statistical significance, although patients predicted to achieve pCR showed a trend toward longer survival.
CONCLUSION: Integrating MRI radiomics with IRSN-23 improved pCR prediction compared with either modality alone and may support treatment planning, while the observed associations with survival outcomes should be interpreted as exploratory.
引用
- Tokuda Y, Suzuki Y, Oda S, et al. A Radiogenomic Model using MRI and Gene Signature to Predict Complete Response in Breast Cancer. Eur J Radiol. 2026. PMID: 41950863. doi:10.1016/j.ejrad.2026.112824
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。