文献速递神经MR

联合DTI特征的结构连接组预测胶质瘤患者术后语言功能下降及其恢复

Structural connectome combining DTI features predicts postoperative language decline and its recovery in glioma patients.
PubMed 文献 · Eur Radiol · 2023-09-22

本研究回顾性纳入81例累及左半球的右利手胶质瘤患者,术前采用联合DTI特征的结构连接组构建机器学习模型,预测术后1周语言功能下降、术后3个月长期语言功能下降及语言功能恢复;三个模型在嵌套交叉验证中的总体准确率分别为0.840、0.790、0.867,AUC分别为0.841、0.778、0.901。

要点

  • 联合DTI特征的结构连接组可预测胶质瘤术后1周语言功能下降,模型总体准确率0.840、AUC 0.841。
  • 该连接组特征可预测术后3个月长期语言功能下降,模型总体准确率0.790、AUC 0.778。
  • 该连接组特征可预测术后语言障碍能否恢复,模型总体准确率0.867、AUC 0.901。

核心结论

  • 作者采用机器学习算法建立模型,预测胶质瘤患者术后语言功能是否下降以及术后语言障碍能否恢复,可能有助于改进治疗策略的制定。

研究设计与样本

  • 回顾性纳入81例累及左半球的右利手胶质瘤患者,术前采用联合DTI特征的结构连接组,以术前及术后1周、3个月的西方失语症成套测验语言功能评估为参考,采用嵌套交叉验证优化模型、评价预测性能并筛选最具预测价值的特征。

诊断标准与关键数值

  • 术后语言功能下降(术后1周)模型最终纳入5个预测特征,嵌套交叉验证总体准确率0.840、总体AUC 0.841。

临床应用与阅片要点

  • 作者认为模型可预测胶质瘤患者术后语言功能下降及能否恢复,可能有助于改进治疗策略的制定;非语言功能下降组或语言功能恢复组中特征的差异可能反映胶质瘤语言功能保护与代偿的结构基础。

局限性

  • 原文未报告局限性。

以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。

英文原文(摘要)· 点击展开

OBJECTIVES: A decline in language function is a common complication after glioma surgery, affecting patients' quality of life and survival. This study predicts the postoperative decline in language function and whether it can be recovered based on the preoperative white matter structural network.

MATERIALS AND METHODS: Eighty-one right-handed patients with glioma involving the left hemisphere were retrospectively included. Their language function was assessed using the Western Aphasia Battery before and 1 week and 3 months after surgery. Structural connectome combining DTI features was selected to predict postoperative language decline and recovery. Nested cross-validation was used to optimize the models, evaluate the prediction performance of the models, and identify the most predictive features.

RESULTS: Five, seven, and seven features were finally selected as the predictive features in each model and used to establish predictive models for postoperative language decline (1 week after surgery), long-term language decline (3 months after surgery), and language recovery, respectively. The overall accuracy of the three models in nested cross-validation and overall area under the receiver operating characteristic curve were 0.840, 0.790, and 0.867, and 0.841, 0.778, and 0.901, respectively.

CONCLUSION: We used machine learning algorithms to establish models to predict whether the language function of glioma patients will decline after surgery and whether postoperative language deficit can recover, which may help improve the development of treatment strategies. The difference in features in the non-language decline or the language recovery group may reflect the structural basis for the protection and compensation of language function in gliomas.

CLINICAL RELEVANCE STATEMENT: Models can predict the postoperative language decline and whether it can recover in glioma patients, possibly improving the development of treatment strategies. The difference in selected features may reflect the structural basis for the protection and compensation of language function.

KEY POINTS: • Structural connectome combining diffusion tensor imaging features predicted glioma patients' language decline after surgery. • Structural connectome combining diffusion tensor imaging features predicted language recovery of glioma patients with postoperative language disorder. • Diffusion tensor imaging and connectome features related to language function changes imply plastic brain regions and connections.

来源:Eur Radiol,查看原文。版权归原作者所有。

引用

  1. Liu Y, Cui M, Gao X, et al. Structural connectome combining DTI features predicts postoperative language decline and its recovery in glioma patients. Eur Radiol. 2023. PMID: 37736802. doi:10.1007/s00330-023-10212-2

供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。