摘要目的:探索人工智能辅助压缩感知（ACS）技术加速鼻咽癌放疗MR模拟定位（MR-sim）的可行性及临床应用价值。方法:前瞻性纳入30例拟于中山大学肿瘤防治中心行根治性放疗的鼻咽癌患者，在3.0 T MR-sim设备行头颈部MR-sim，分别采用ACS技术和并行采集（PI）技术，采集每位患者的横断面T 1WI、T 2WI、增强T 1WI及脂肪抑制增强T 1WI图像。采用配对样本 t检验或秩和检验，对比两种采集技术获取MR-sim图像的扫描时间、图像信噪比（SNR）、对比噪声比（CNR）；基于利克特量表5级评分法，对肿瘤病灶显示情况、病灶边缘清晰度、图像伪影及图像整体质量进行评分，并采用卡方检验对比2种技术的主观图像质量评分。分析基于ACS获取的MR-sim图像和基于PI技术获取的MR-sim图像，与CT模拟定位图像融合后勾画肿瘤靶区，并采用 Dice相似性系数（DSC）对肿瘤靶区勾画一致性进行评价。 结果:无论对于单个序列还是整体序列扫描，基于ACS技术的MR-sim采集时间都显著短于PI技术（ P<0.001），其中基于ACS的MR-sim总体采集时间为（378.60±17.07）s，基于PI技术的为（694.93±17.07）s，时间节约45.52%。基于ACS技术和基于PI技术获取的MR-sim图像SNR、CNR及肿瘤病灶识别、病灶边缘清晰度、图像伪影、总体图像质量评分均基本一致（ P>0.05）。基于ACS获取的MR-sim图像和基于PI技术获取的MR-sim图像，与CT模拟定位图像融合后获取的肿瘤靶区基本一致（ P>0.05），原发肿瘤和颈部转移淋巴结平均DSC均接近于1。 结论:与传统的MR加速成像技术（PI技术）相比，ACS技术可以在不牺牲图像质量和不影响肿瘤靶区勾画的情况下，加速鼻咽癌MR-sim。

abstractsObjective:To investigate the feasibility and clinical value of artificial intelligence-assisted compressed sensing (ACS) technology in accelerating MR simulation (MR-sim) for radiotherapy of nasopharyngeal carcinoma (NPC).Methods:Thirty patients with NPC scheduled to receive radical radiotherapy at Sun Yat-sen University Cancer Center were prospectively enrolled. All patients underwent head and neck MR-sim on a 3.0 T scanner, with axial T 1 weighted imaging (WI), T 2WI, contrast-enhanced T 1WI, and fat-suppressed contrast-enhanced T 1WI images acquired using both ACS and parallel imaging (PI) techniques. Paired-sample t tests or rank-sum tests were used to compare scan time, signal-to-noise ratio (SNR), and contrast-to-noise ratio (CNR) of MR-sim images between the two techniques. A 5-point Likert scale was applied to evaluate tumor lesion visualization, lesion margin clarity, artifacts, and overall image quality, with chi-square tests used to compare subjective image quality scores between the two techniques. Tumor target volumes were delineated on MR-sim images obtained by both ACS and PI techniques after fusion with CT simulation images, and consistency was assessed using the Dice similarity coefficient (DSC). Results:For both individual sequences and overall protocols, ACS significantly reduced MR-sim acquisition time compared with PI ( P < 0.001). The total acquisition time with ACS was (378.60±17.07) s versus (694.93±17.07) s with PI, representing a 45.52% time reduction. SNR, CNR, tumor lesion identification, margin clarity, artifacts, and overall image quality scores of MR-sim images did not differ significantly between ACS and PI ( P > 0.05). Tumor target volumes delineated from ACS- and PI-based MR-sim images showed high consistency after fusion with CT simulation images ( P > 0.05), with mean DSC values of primary tumors and metastatic cervical lymph nodes approaching 1. Conclusion:Compared with conventional MR acceleration methods (PI), ACS enables faster MR-sim acquisition in NPC without compromising image quality or the accuracy of tumor target delineation.