BackgroundVascular cognitive impairment (VCI) is a major cause of cognitive impairment in the elderly and a co-factor in the development and progression of most neurodegenerative diseases. With the continuing development of neuroimaging, multiple markers can be combined to provide richer biological information, but little is known about their diagnostic value in VCI.MethodsA total of 83 subjects participated in our study, including 32 patients with vascular cognitive impairment with no dementia (VCIND), 21 patients with vascular dementia (VD), and 30 normal controls (NC). We utilized resting-state quantitative electroencephalography (qEEG) power spectra, structural magnetic resonance imaging (sMRI) for feature screening, and combined them with support vector machines to predict VCI patients at different disease stages.ResultsThe classification performance of sMRI outperformed qEEG when distinguishing VD from NC (AUC of 0.90 vs. 0,82), and sMRI also outperformed qEEG when distinguishing VD from VCIND (AUC of 0.8 vs. 0,0.64), but both underperformed when distinguishing VCIND from NC (AUC of 0.58 vs. 0.56). In contrast, the joint model based on qEEG and sMRI features showed relatively good classification accuracy (AUC of 0.72) to discriminate VCIND from NC, higher than that of either qEEG or sMRI alone.ConclusionPatients at varying stages of VCI exhibit diverse levels of brain structure and neurophysiological abnormalities. EEG serves as an affordable and convenient diagnostic means to differentiate between different VCI stages. A machine learning model that utilizes EEG and sMRI as composite markers is highly valuable in distinguishing diverse VCI stages and in individually tailoring the diagnosis.

中文翻译：

---

基于脑电图和结构磁共振成像的机器学习可以预测血管性认知障碍的不同阶段

背景血管性认知障碍（VCI）是老年人认知障碍的主要原因，也是大多数神经退行性疾病发生和进展的辅助因素。随着神经影像学的不断发展，多种标志物可以结合起来提供更丰富的生物学信息，但对其在VCI中的诊断价值知之甚少。 方法共有83名受试者参与了我们的研究，其中包括32名患有血管性认知障碍但不伴有痴呆的患者（ VCIND）、21 名血管性痴呆（VD）患者和 30 名正常对照（NC）。我们利用静息态定量脑电图（qEEG）功率谱、结构磁共振成像（sMRI）进行特征筛选，并结合支持向量机来预测不同疾病阶段的VCI患者。结果在区分VD时，sMRI的分类性能优于qEEG与 NC 的比较（AUC 为 0.90 vs. 0.82），sMRI 在区分 VD 与 VCIND 时也优于 qEEG（AUC 为 0.8 vs. 0.0.64），但在区分 VCIND 与 NC 时两者均表现不佳（AUC 为 0.58 vs. 0.56） 。相比之下，基于qEEG和sMRI特征的联合模型在区分VCIND和NC方面表现出相对较好的分类准确性（AUC为0.72），高于单独的qEEG或sMRI。结论VCI不同阶段的患者表现出不同水平的大脑结构和神经生理学异常。脑电图是一种经济且方便的诊断手段，可以区分不同的 VCI 阶段。利用 EEG 和 sMRI 作为复合标记的机器学习模型对于区分不同的 VCI 阶段和单独定制诊断非常有价值。