基于治疗前磁共振弥散加权成像（DWI）信号值构建局部进展期胰腺癌（LAPC）续贯放化疗生存获益预测模型。

回顾性分析2015年1月至2017年12月在海军军医大学第一附属医院接受立体定向放疗（SBRT）的39例LAPC患者临床影像学资料。患者均签署知情同意书。其中男26例，女13例；年龄33～80岁，中位年龄64岁。胰头癌33例，胰体尾癌6例。30例患者SBRT放疗后续贯替吉奥（S-1）化疗。SBRT放疗前均行多b值下DWI序列成像检查，成像时采用11个b值（0、25、50、75、100、150、200、400、600、800、1 000 s/mm²），得到11帧不同b值下的重叠DWI图像。采用Cox回归分析不同b值下的感兴趣区域（ROI）平均信号值（SI）与总体生存期（OS）之间的相关性，筛选独立风险因素，建立列线图预测模型，预测拟行续贯放化疗LAPC患者的生存获益情况。预测误差曲线（PEC）分析通过计算Brier综合分数（IBS）来评估预测误差。

随访时间87～1 095 d，中位随访时间353 d；随访期间38例死亡，1例存活。患者OS为84～1 095 d，中位OS为352 d，1年生存率46%。Cox回归分析显示，SI₁₀₀（HR=0.997，95%CI：0.996～0.998）、SI₆₀₀（HR=0.996，95%CI：0.993～0.998）、SI₈₀₀ （HR=0.994，95%CI：0.991～0.997）和S₁₀₀₀（HR=0.993，95%CI：0.989～0.996）是OS的独立预测因素（P<0.05）。PEC分析显示，预测模型1 （SI₁₀₀ + S-1，C-index 0.712）、模型2 （SI₆₀₀ +年龄+S-1，C-index 0.731）、模型3 （SI₈₀₀ +年龄+ S-1，C-index 0.736）、模型4 （SI₁₀₀₀ +年龄+ S-1，C-index 0.732）均较模型内其他组合预测误差小，IBS分别为0.134、0.133、0.130和0.133，并构建相应的列线图预测模型。线性校准图显示预测效果及实际观察效应之间具有很高的一致性，进一步证实列线图预测模型的可靠性。

基于放化疗前DWI序列信号值的列线图预测模型能够对拟进行放化疗LAPC患者的生存期获益进行预测，为临床医师制定个体化治疗决策提供依据，避免过度医疗。

To establish a prediction model for survival benefits of locally advanced pancreatic cancer (LAPC) patients after sequential radiochemotherapy based on the signal value of diffusion-weighted imaging (DWI) before treatment.

Clinical imaging data of 39 patients with LAPC who received stereotactic body radiotherapy (SBRT) in the First Affiliated Hospital of Naval Medical University from January 2015 to December 2017 were retrospectively analyzed. The informed consents of all patients were obtained and the local ethical committee approval was received. Among them, 26 patients were male and 13 female, aged from 33 to 80 years, with a median age of 64 years. 33 patients were diagnosed with pancreatic head cancer and 6 cases of pancreatic body and tail cancer. 30 patients were treated with SBRT followed by S-1 chemotherapy. DWI sequence imaging with multiple b values was performed before SBRT. 11 b values (0, 25, 50, 75, 100, 150, 200, 400, 600, 800, 1 000 s/mm²) were adopted to obtain 11 overlapping DWI images with different b values. Cox regression model was used to analyze the correlation between the average signal intensity (SI) of the region of interests (ROI) and overall survival (OS) under different b values, to screen the independent risk factors, and to establish a nomogram prediction model, thereby predicting the survival benefits of patients with LAPC scheduled to undergo sequential radiochemotherapy. Prediction error curve (PEC) analysis was utilized to assess the prediction error by calculating integrated Brier score (IBS).

The follow-up time was ranged from 87 to 1 095 d, with a median of 353 d. During follow-up, 38 patients died and 1 survived. The OS of all patients was ranged from 84 to 1 095 d, with a median OS of 352 d, and the 1-year survival rate was 46%. Cox regression analysis showed that SI₁₀₀ (HR=0.997, 95%CI: 0.996-0.998), SI₆₀₀ (HR=0.996, 95%CI: 0.993-0.998), SI₈₀₀ (HR=0.994, 95%CI: 0.991-0.997) and SI₁₀₀₀ (HR=0.993, 95%CI: 0.989-0.996) were the independent predictive factors of OS (P<0.05). PEC analysis revealed that prediction model 1 (SI₁₀₀ + S-1, C-index 0.712), model 2 (SI₆₀₀ + age + S-1, C-index 0.731), model 3 (SI₈₀₀ + age + S-1, C-index 0.736) and model 4 (SI₁₀₀₀ + age + S-1, C-index 0.732) had smaller errors than other combinations within each model, and the IBS was 0.134, 0.133, 0.130 and 0.133, respectively. Corresponding nomogram prediction models were also established. Linear calibration chart indicated a high degree of consistency between the prediction effect and the actual effect, which further confirmed the reliability of the nomogram prediction model.

The nomogram prediction model based on the signal value of DWI sequence before radiochemotherapy can predict the survival benefits of LAPC patients scheduled to undergo radiochemtherapy, providing evidence for clinicians to make individualized treatment decisions and avoid overtreatment.