Yang Wen, Nie Qilin, Sun Yujie, Zou Danrong, Tang Jinmo, Wang Min
Xiamen Hospital of Traditional Chinese Medicine, Xiamen, China.
Front Physiol. 2023 Jul 20;14:1225636. doi: 10.3389/fphys.2023.1225636. eCollection 2023.
Atherosclerosis is a chronic vascular disease that poses a significant threat to human health. Common diagnostic methods mainly rely on active screening, which often misses the opportunity for early detection. To overcome this problem, this paper presents a novel medical ambient intelligence system for the early detection of atherosclerosis by leveraging clinical data from medical records. The system architecture includes clinical data extraction, transformation, normalization, feature selection, medical ambient computation, and predictive generation. However, the heterogeneity of examination items from different patients can degrade prediction performance. To enhance prediction performance, the "SEcond-order Classifier (SEC)" is proposed to undertake the medical ambient computation task. The first-order component and second-order cross-feature component are then consolidated and applied to the chosen feature matrix to learn the associations between the physical examination data, respectively. The prediction is lastly produced by aggregating the representations. Extensive experimental results reveal that the proposed method's diagnostic prediction performance is superior to other state-of-the-art methods. Specifically, the Vitamin B12 indicator exhibits the strongest correlation with the early stage of atherosclerosis, while several known relevant biomarkers also demonstrate significant correlation in experimental data. The method proposed in this paper is a standalone tool, and its source code will be released in the future.
动脉粥样硬化是一种对人类健康构成重大威胁的慢性血管疾病。常见的诊断方法主要依赖于主动筛查，这往往会错过早期检测的机会。为了克服这一问题，本文提出了一种新颖的医学环境智能系统，通过利用病历中的临床数据来早期检测动脉粥样硬化。该系统架构包括临床数据提取、转换、归一化、特征选择、医学环境计算和预测生成。然而，不同患者检查项目的异质性会降低预测性能。为了提高预测性能，提出了“二阶分类器（SEC）”来承担医学环境计算任务。然后将一阶分量和二阶交叉特征分量合并，并应用于所选特征矩阵，以分别学习体检数据之间的关联。最后通过聚合表示来产生预测。大量实验结果表明，所提出方法的诊断预测性能优于其他现有方法。具体而言，维生素B12指标与动脉粥样硬化早期的相关性最强，而一些已知的相关生物标志物在实验数据中也显示出显著相关性。本文提出的方法是一个独立工具，其源代码将在未来发布。
