目的：探讨深度学习图像重建（deep learning image reconstruction，DLIR）算法应用于术前双能CT评估胃癌T分期中的价值。方法：回顾分析本院2022年1月至2022年2月期间，经术后病理确诊的45例胃癌患者的术前双能CT检查资料，其中静脉期双能扫描原始数据，以标准重建核分别使用滤波反投影重建（filtered back projection，FBP），自适应统计迭代重建（adaptive statistical iterative reconstruction-V，Asir-V）权重50%（AV-50）和权重100%（AV-100），和DLIR中档（deep learning image reconstruction - mid-range ，DLIR-M）算法，重建1.25 mm层厚的50 keV能级虚拟单能图像，随后由具有5年和10年胃肠道肿瘤诊断经验的2名放射科医师协商进行胃癌T分期诊断。45例患者中，术后病理学诊断为早期胃癌（T_1a~T_1b期）者22例，进展期胃癌（T₂~T₃期）者23例。以病理结果为金标准，采用受试者操作特征曲线的曲线下面积（area under curve,AUC）比较不同重建算法图像评估胃癌T分期的效能。结果：双能CT检查FBP、AV-50、AV-100和DLIR-M算法在术前判断胃癌T分的受试者操作特征曲线的AUC分别为0.638、0.667、0.577和0.867。DLIR-M图像的AUC高于FBP（P=0.049 8）、A-V-50（P=0.047 7）、AV-100（P=0.012 3）。结论：与传统的FBP、Asir-V等CT重建算法相比，DLIR-M算法能进一步提高双能CT检查在术前判断胃癌T分期的准确率，对临床治疗方案选择更具指导意义。

Objective: To investigate the value of deep learning image reconstruction (DLIR) algorithm in dual-energy CT scanning for preoperative T staging of gastric cancer. Methods: Data from preoperative dual-energy CT of 45 patients with pathology-confirmed gastric cancer during January 2022 to February 2022 were retrospectively analyzed. The raw data of dual-energy scanning in venous phase were reconstructed by filtered back projection (FBP), adaptive statistical iterative reconstruction-V with a weight of 50% (AV-50) and of 100% (AV-100), and deep learning image reconstruction-mid-range (DLIR-M) algorithms. Then these images were used to reconstruct a 50 keV level virtual mono-energy image with a 1.25mm layer thickness. Images were reviewed by two radiologists with 5 and 10 years of experience in T staging of gastrointestinal tumors. It revealed that 22 cases were diagnosed pathologically as having early gastric cancer (T_1a-T_1b) and 23 cases as having advanced gastric cancer (T₂-T₃). Diagnostic accuracy of different reconstruction algorithms for T staging of gastric cancer were calculated using area under the receiver operating curve (AUC) ，with pathology results as the golden standard. Results: The AUC of the reconstructed dual-energy CT images based on FBP, AV-50, AV-100 and DLIR-M algorithms were 0.638, 0.667, 0.577 and 0.867, respectively. The AUC of DLIR-M images was significantly higher than those of FBP (P=0.0498), AV-50 (P=0.0477) and AV-100 (P=0.0123) images. Conclusions: Compared with traditional reconstruction algorithms of FBP and Asir-V，DLIR algorithm may further improve the accuracy of dual-energy CT scanning in preoperative T staging of gastric cancer. DLIR-M is significant for treatment decision-making.