Abstract: Objective    To assess the differences in velopharyngeal structures between adult patients with velopharyngeal insufficiency（VPI）and normal volunteers using dynamic magnetic resonance imaging（MRI），and to develop a surgical decision-making model. Methods    A total of 38 adult patients with VPI（study group）and 30 volunteers with normal velopharyngeal closure（control group），who were treated at Peking Union Medical College Hospital from September 2021 to January 2025，were enrolled in this study. MRI was used to measure velopharyngeal structural parameters at rest and during phonation. Differences between the two groups were compared. Parameters with significant differences were incorporated into a multiple regression model to construct a surgical decision-making model. The accuracy of the model was evaluated using 4-fold cross-validation and follow-up data. Results    At rest，the angle between the bilateral levator veli palatini（LVP）muscles and the cranial base，as well as the velopharyngeal area，were significantly greater in the study group than in the control group. During phonation，the right LVP thickness，velar elevation ratio，velopharyngeal closure ratio，velar length，velar thickness，effective velar length/（effective velar length plus pharyngeal depth），palatal length，the angle between the trajectory of the LVP insertion point into the soft palate and the sella-nasion（SN） plane，and the displacement distance of the LVP insertion point into the soft palate were significantly smaller in the study group than in the control group，whereas the angle between the left LVP and the cranial base，the velopharyngeal area，and pharyngeal depth were significantly greater in the study group. All differences were statistically significant（P < 0.05）. No statistically significant differences were observed between the two groups in bilateral LVP length and velar height at rest and during phonation，bilateral LVP thickness and velar thickness at rest，or left LVP thickness，the angle between the right LVP and the cranial base，and effective velar length during phonation（P > 0.05）. Based on the discriminating variables and considering multicollinearity among the variables，five indices，namely，velar elevation ratio during phonation（X1），velopharyngeal closure ratio（X2），the angle between the trajectory of the LVP insertion point into the soft palate and the SN plane（X3），the displacement distance of the LVP insertion point into the soft palate（X4），and effective velar length/（effective velar length plus pharyngeal depth）（X5），were ultimately included in a multiple regression analysis. The resulting surgical decision model was：Y = 2.161 + 1.718 X1 - 0.381 X2 - 1.025 X3 - 0.681 X4 - 1.279 X5（R2 = 0.504，P = 0.006）. A Y value tending toward 1 suggested that surgical intervention was indicated，whereas a Y value tending toward 0 suggested that speech therapy was sufficient and surgery was not required. This VPI surgical decision model was validated using dynamic MRI data from 30 newly recruited patients with VPI，and the results demonstrated a significant correlation with expert opinion（χ2 = 12.805，P < 0.001）. Conclusion     Based on the assessment indices of velopharyngeal structures derived from dynamic MRI，this study has successfully developed a predictive model for surgical decisionmaking in VPI. The model can provide an objective and accurate evaluation system for the clinical formulation of treatment strategies for VPI.

Key words: velopharyngeal insufficiency, dynamic magnetic resonance imaging, velopharyngeal structures, surgical decision-making model, multiple regression analysis

摘要： 目的    使用动态磁共振成像（magnetic resonance imaging，MRI）评估腭咽闭合不全（velopharyngeal insufficiency，VPI）成人患者与正常志愿者的腭咽结构差异，并构建手术决策模型。方法    选取2021年9月至2025年1月于北京协和医院口腔科就诊的38例VPI成人患者（研究组）和招募的30名腭咽闭合正常的志愿者（对照组）作为研究对象，通过动态MRI测量静息及发音时的腭咽结构指标并比较组间差异。将差异指标纳入多元回归模型构建手术决策模型，并利用4折交叉验证和随访数据评估模型准确性。结果    研究组静息时双侧腭帆提肌（levator veli palatini，LVP）与颅底所成角度、咽腔面积明显大于对照组；发音时右侧LVP厚度、软腭抬高比率、咽腔闭合比率、软腭长度、软腭厚度、有效软腭长度/（有效软腭长度+咽腔深度）、硬腭长度、LVP进入软腭的点的运动轨迹与前颅底（sella-nasion，SN）平面所成角度、LVP进入软腭的点的运动轨迹距离明显小于对照组，左侧LVP与颅底所成角度、咽腔面积、咽腔深度明显大于对照组，差异均有统计学意义（均P < 0.05）。研究组和对照组静息时和发音时双侧LVP长度、软腭高度，静息时双侧LVP厚度、软腭厚度，以及发音时左侧LVP厚度、右侧LVP与颅底所成角度、有效软腭长度比较，差异均无统计学意义（均P > 0.05）。基于差异指标并考虑变量间多重共线性问题，最终将发音时软腭抬高比率（X1）、咽腔闭合比率（X2）、LVP进入软腭的点的运动轨迹与SN平面所成角度（X3）、LVP进入软腭的点的运动轨迹距离（X4）、有效软腭长度/（有效软腭长度+咽腔深度）（X5）纳入多元回归分析，得到手术决策模型为Y = 2.161 + 1.718 X1 - 0.381 X2 - 1.025 X3 - 0.681 X4 - 1.279 X5（R2 = 0.504，P = 0.006）。Y值趋向于1，则提示需手术干预；Y值趋向于0，则提示需语音训练，无需手术干预。通过新招募的30例VPI患者的动态MRI结果来验证该VPI手术决策模型，结果表明其与专家意见存在显著关联（χ2 = 12.805，P < 0.001）。结论    研究基于动态MRI中的腭咽结构评估指标成功构建了VPI手术决策预测模型，可为临床制定VPI的治疗策略提供一种客观且准确的评价体系。

关键词: 腭咽闭合不全, 动态磁共振成像, 腭咽结构, 手术决策模型, 多元回归分析