Department of Rehabilitation Medicine, The Affiliated Suzhou Hospital of Nanjing Medical University, Suzhou 215008, China.
Department of Radiology, The Affiliated Suzhou Hospital of Nanjing Medical University, Suzhou 215008, China.
Brain Res Bull. 2024 Oct 15;217:111074. doi: 10.1016/j.brainresbull.2024.111074. Epub 2024 Sep 7.
To construct relatively objective, atlas-based multivariate models for predicting early aphasia severity after stroke, using structural magnetic resonance imaging.
We analyzed the clinical and imaging data of 46 patients with post-stroke aphasia. The aphasia severity was identified with a Western Aphasia Battery Aphasia Quotient. The assessments of stroke lesions were indicated by the lesion load of both the cortical language areas (Areas-LL) and four white matter tracts (i.e., the superior longitudinal fasciculus, SLF-LL; the inferior frontal occipital fasciculi, IFOF-LL; the inferior longitudinal, ILF-LL; and the uncinate fasciculi, UF-LL) extracted from human brain atlas. Correlation analyses and multiple linear regression analyses were conducted to evaluate the correlations between demographic, stroke- and lesion-related variables and aphasia severity. The predictive models were then established according to the identified significant variables. Finally, the receiver operating characteristic (ROC) curve was utilized to assess the accuracy of the predictive models.
The variables including Areas-LL, the SLF-LL, and the IFOF-LL were significantly negatively associated with aphasia severity (p < 0.05). In multiple linear regression analyses, these variables accounted for 59.4 % of the variance (p < 0.05). The ROC curve analyses yielded the validated area under the curve (AUC) 0.84 both for Areas-LL and SLF-LL and 0.76 for IFOF-LL, indicating good predictive performance (p < 0.01). Adding the combination of SLF-LL and IFOF-LL to this model increased the explained variance to 62.6 % and the AUC to 0.92.
The application of atlas-based multimodal lesion assessment may help predict the aphasia severity after stroke, which needs to be further validated and generalized for the prediction of more outcome measures in populations with various brain injuries.
使用结构磁共振成像构建相对客观的基于图谱的多变量模型，以预测中风后的早期失语症严重程度。
我们分析了 46 例中风后失语症患者的临床和影像数据。使用西方失语症成套测验失语商评估失语症严重程度。使用从人脑图谱中提取的皮质语言区病变负荷（Areas-LL）和四个白质束（即上纵束、SLF-LL；下额枕束、IFOF-LL；下纵束、ILF-LL；和钩束、UF-LL）评估中风病变。进行相关性分析和多元线性回归分析，以评估人口统计学、中风和病变相关变量与失语症严重程度之间的相关性。然后根据确定的显著变量建立预测模型。最后，利用受试者工作特征（ROC）曲线评估预测模型的准确性。
Areas-LL、SLF-LL 和 IFOF-LL 等变量与失语症严重程度呈显著负相关（p<0.05）。多元线性回归分析表明，这些变量解释了 59.4%的方差（p<0.05）。ROC 曲线分析显示，Areas-LL 和 SLF-LL 的验证曲线下面积（AUC）分别为 0.84，IFOF-LL 的 AUC 为 0.76，表明预测性能良好（p<0.01）。将 SLF-LL 和 IFOF-LL 的组合添加到该模型中，可将解释的方差增加到 62.6%，AUC 增加到 0.92。
基于图谱的多模态病变评估的应用可能有助于预测中风后的失语症严重程度，需要进一步验证和推广，以预测各种脑损伤人群的更多预后指标。
