目的:探讨基于MRI影像组学的预测模型在程序性死亡蛋白-1（PD-1）抗体联合全程新辅助放化疗后中低位直肠癌病理完全缓解（pCR）中的预测价值。方法:采用临床诊断性试验研究方法，收集2019年1月至2021年9月期间于北京大学肿瘤医院收治的PD-1抗体联合全程新辅助放化疗并最终行根治性手术的38例中低位直肠癌患者的临床病理资料和影像资料；男23例，女15例；中位年龄68（47~79）岁；有13例为pCR，25例为非pCR。将38例患者分层随机分为训练组（26例）和测试组（12例）进行建模。所有患者治疗前均行直肠MRI检查，收集所有患者的临床特征、影像特征和影像组学特征，并构建临床特征模型和影像组学模型。绘制各个模型的受试者工作特征（ROC）曲线，并以曲线下面积、准确度、灵敏度、特异度、阳性预测值和阴性预测值等来评价所构建模型。结果:两组患者的年龄、性别、肿瘤原发位置和术后病理情况比较，差异均无统计学意义（均n P>0.05）。本研究在38例中低位直肠癌患者中，从每个模态下感兴趣区域（ROI）提取了41个特征，包括9个一阶特征，24个灰度共生矩阵特征和8个形状特征。进一步分别从新辅助治疗前后弥散加权成像（DWI）和T2加权成像（T2WI）的每一个图像模态中提取41个特征，共纳入164特征。经过n t检验和相关系数的筛选后，仅有4个特征得到保留。经过LASSO交叉验证后，仅有治疗前基线DWI图像的一阶偏度和治疗前基线T2WI图像中的体积这两个特征得到保留。应用这两个特征建立的预测模型在训练组和测试组的曲线下面积、灵敏度、特异度、阳性预测值和阴性预测值分别为0.856和0.844，77.8%和100.0%，88.2%和75.0%，77.8%和66.7%，88.2%和100.0%。影像组学模型的决策曲线显示，采用该预测模型预测pCR的策略要优于将所有患者都看作pCR的策略，也优于将所有患者看作非pCR的策略。n 结论:基于MRI影像组学构建PD-1抗体联合全程新辅助放化疗治疗直肠癌pCR预测模型，有潜力用于临床筛选或可免于根治性手术的直肠癌患者。“,”Objective:To construct a prediction model of pathologic complete response (pCR) in locally advanced rectal cancer patients who received programmed cell death protein-1 (PD-1) antibody and total neoadjuvant chemoradiotherapy by using radiomics based on MR imaging data and to investigate its predictive value.Methods:A clinical diagnostic test study was carried out. Clinicopathalogical and radiological data of 38 patients with middle-low rectal cancer who received PD-1 antibody combined with total neoadjuvant chemoradiotherapy and underwent TME surgery from January 2019 to September 2021 in our hospital were retrospectively collected. Among 38 patients, 23 were males and 15 were females with a median age of 68 (47-79) years and 13 (34.2%) a chieved pCR. These 38 patients were stratified and randomly divided into the training group (n n=26) and test group (n n=12) for modeling. All the patients underwent rectal MRI before treatment. The clinical, imaging and radiomics features of all the patients were collected, and the clinical feature model and radiomics model were constructed. The receiver operating characteristic (ROC) curves of each model were drawn, and the constructed model was evaluated through the area under the curve (AUC), accuracy, sensitivity, specificity, positive predictive value and negative predictive value.n Results:There were no significant differences in age, gender, primary location of tumor and postoperative pathology between the two groups (all n P>0.05). Forty-one features were extracted from region of interest in each modality, including 9 first-order features, 24 gray level co-occurrence matrix features and 8 shape features. From 38 patients, 41 features were extracted from each imaging modality of baseline and preoperative DWI and T2WI images, totally 164 features. Only 4 features were preserved after correlation analysis between each pair of features and t-test between pCR and non-pCR subjects. After LASSO cross validation, only the first-order skewness of the baseline DWI image before treatment and the volume in the baseline T2WI image before treatment were retained. The area under the curve, sensitivity, specificity, positive and negative predictive values of the prediction model established by applying these two features in the training group and the test group were 0.856 and 0.844, 77.8% and 100.0%, 88.2% and 75.0%, 77.8% and 66.7%, 88.2% and 100.0%, respectively. The decision curve analysis of the radiomics model showed that the strategy of this model in predicting pCR was better than that in treating all the patients as pCR and that in treating all the patients as non-pCR.n Conclusion:The pCR prediction model for rectal cancer patients receiving PD-1 antibody combined with total neoadjuvant radiochemotherapy based on MRI radiomics has the potential to be used in clinical screening or rectal cancer patients who can be spared from radical surgery.