基于MRI影像预测局部进展期直肠癌新辅助治疗疗效预测模型

Predictive Model for Neoadjuvant Therapy Response in Locally Advanced Rectal Cancer Based on MRI Imaging

摘要: 目的：构建并验证基于多参数MRI影像组学的预测模型，用于术前预测局部进展期直肠癌(LARC)患者新辅助放化疗(nCRT)后的病理完全缓解(pCR)状态，为“观察–等待”治疗策略的精准实施提供循证依据。方法：回顾性纳入2018年1月至2023年6月期间接受nCRT及根治性手术的LARC患者102例，按7:3比例随机分为训练集(n = 71)与验证集(n = 31)。基于治疗前3.0 T多参数MRI (含高分辨率T2WI、DWI及DCE序列)提取影像组学特征，经ICC稳定性筛选、相关性分析及LASSO回归降维后构建影像组学标签(Rad-score)。分别建立临床模型、影像组学模型及联合模型，采用受试者工作特征曲线下面积(AUC)评估模型性能，Hosmer-Lemeshow检验评价校准度。结果：经三级降维后最终保留12个最优影像组学特征构建Rad-score。验证集中，单纯临床模型AUC为0.77 (95% CI: 0.62~0.89)，联合模型AUC达0.86 (95% CI: 0.80~0.92)，显著优于单一模型。校准曲线显示联合模型预测概率与实际观测一致性良好。结论：基于高分辨率MRI的影像组学联合模型可在术前有效识别pCR患者，具备指导直肠癌个体化治疗决策的临床应用价值，但尚需外部多中心验证及自动分割技术优化。

Abstract: Objective: To develop and validate a multiparametric MRI-based radiomics model for preoperative prediction of pathological complete response (pCR) following neoadjuvant chemoradiotherapy (nCRT) in patients with locally advanced rectal cancer (LARC), thereby providing evidence-based support for the precise implementation of the “watch-and-wait” strategy. Methods: A retrospective cohort of 102 LARC patients who underwent nCRT and radical surgery between January 2018 and June 2023 was enrolled and randomly divided into training (n = 71) and validation (n = 31) sets at a 7:3 ratio. Radiomics features were extracted from pretreatment 3.0 T multiparametric MRI (including high-resolution T2WI, DWI, and DCE sequences) and subjected to a three-step dimensionality reduction involving ICC-based stability screening, correlation analysis, and LASSO regression to construct a radiomics signature (Rad-score). Clinical, radiomics, and combined models were developed respectively. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), with DeLong test for inter-model comparison and Hosmer-Lemeshow test for calibration assessment. Results: Twelve optimal radiomics features were retained after three-step dimensionality reduction to construct the Rad-score. In the validation cohort, the clinical model achieved an AUC of 0.77 (95% CI: 0.62~0.89), while the combined model demonstrated superior performance with an AUC of 0.86 (95% CI: 0.80~0.92). Calibration curves indicated good agreement between predicted probabilities and actual observations for the combined model. Conclusion: The high-resolution MRI-based radiomics combined model enables effective preoperative identification of pCR patients and holds clinical applicability for guiding individualized treatment decisions in rectal cancer, though external multicenter validation and automated segmentation optimization remain warranted.