基于机器学习的睡眠剥夺注意力易损性分类研究

目的:旨在寻找可以对睡眠剥夺后注意力易损与耐受个体进行准确区分的白质纤维束。方法:借助弥散张量成像技术获取各向异性分数、轴向扩散系数、径向扩散系数及平均扩散系数等反映白质弥散特性的特征参数，使用支持向量机分类算法构建睡眠剥夺易损性分类模型；采用准确性、敏感性、特异性、阳性预测值和阴性预测值等指标评价分类模型的性能表现；采用置换检验评估分类模型的显著性。结果:与只采用单一类型特征相比，使用组合特征构建的分类器表现性能最佳，其准确性、敏感性、特异性、阳性预测值、阴性预测值及曲线下面积分别为83.67%、80.00%、87.50%、86.96%、80.77%、88.67%。在组合特征构建的分类模型中对分类贡献较大的白质纤维束主要包括放射冠、内囊前肢、丘脑后辐射及皮质脊髓束等投射纤维、上纵束和扣带等联络纤维以及胼胝体和穹窿联合等联合纤维。结论:特定脑区间白质纤维束的微观结构特性可以作为区分睡眠剥夺后注意力易损与耐受个体的影像学标志物。

Classification of attention vulnerability to sleep deprivation based on machine learning

Abstract: Objective To find white matter fiber tracts that can accurately distinguish between individuals who are vulnerable and resistant to sleep deprivation. Methods The characteristic parameters such as fractional anisotropy, axial diffusivity, radial diffusivity and mean diffusivity which reflect the diffusion characteristics of white matter were obtained using diffusion tensor imaging technology. The support vector machine algorithm was used to construct sleep deprivation vulnerability classification model. Finally, the performance of the classification model was assessed by accuracy, sensitivity, specificity, positive predictive value and negative predictive value and the significance of the classification model was evaluated by permutation test. Results Compared with the classifier constructed with a single type of feature, the combined features-based classifier achieved the best classification performance, with the accuracy, sensitivity, specificity, positive predictive value, negative predictive value and AUC of 83.67%, 80.00%, 87.50%, 86.96%, 80.77% and 88.67%, respectively. In the combined features-based classification model, the most discriminative white matter fiber tracts that contributed to the classification mainly included projection fibers (corona radiata, anterior limb of internal capsule, posterior thalamic radiation and corticospinal tract, etc), association fibers (superior longitudinal fasciculus and cingulum, etc), and commissural fibers (corpus callosum and fornix, etc). Conclusion The microstructure of specific white matter fiber tracts can be used as potential imaging markers to distinguish between individuals vulnerable and resistant to sleep deprivation.