Scholarlink

<p>The absence of clinically validated biomarkers and objective diagnostic protocols hinders the accurate and effective diagnosis of depression. Although machine learning has been increasingly explored in psychiatric diagnosis, there remains a pressing need to develop a reliable tool that integrates multimodal data—such as clinical features, cognitive functions, electroencephalographic microstates, and inflammation-associated proteins—to improve diagnostic accuracy and objectivity. One hundred and fifteen patients with depression and 66 healthy controls were included in this study, and data on their clinical characteristics, cognitive function, electroencephalographic microstates, and serum inflammation-related proteins were collected. The baseline depression group was followed up after 4 weeks of clinical treatment, with 56 participants completing the follow-up survey, which mirrored the baseline survey. The depression baseline and healthy control groups were designated as the training set, while the follow-up and healthy control groups served as the validation set. Six classical machine learning algorithms—Decision Tree, Random Forest, XGBoost, LightGBM, k-Nearest Neighbor, and Support Vector Machine—were employed to train the diagnostic prediction model using the training set. The model was then validated with the validation set to identify the optimal depression diagnostic prediction model. The results of the study showed significant differences in clinical characteristics, cognitive function, EEG microstates, and serum levels of inflammation-related proteins in patients with major depressive disorder compared with healthy controls. In the model evaluation, the k-nearest neighbor model performed the best, with an accuracy of 95.08% for multimodal diagnosis, an F1 score of 0.9545, and an AUC value of 0.9969. In order of feature importance were IL-8, IL-18, ISI, MMP-8, CD40, CASP-8, visuospatial/constructional, and mean duration of EEG microstate D. A multimodal, multi-indicator model (incorporating L-8, IL-18, ISI, MMP-8, CD40, CASP-8, visuospatial/constructional abilities, and mean duration of EEG microstate D) has the potential to enhance the accuracy and objectivity of clinical depression diagnoses.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Prediction of depressive episodes based on clinical features, cognitive characteristics, inflammation-related proteins, and EEG data

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月24日

查看全文
  • Wenxi Sun,
  • Haidong Yang,
  • Chao Sun,
  • Qing Tian,
  • Peng Chen,
  • Shiting Yuan,
  • Xueying Zhang,
  • Jin Li,
  • Xiaobin Zhang

摘要

The absence of clinically validated biomarkers and objective diagnostic protocols hinders the accurate and effective diagnosis of depression. Although machine learning has been increasingly explored in psychiatric diagnosis, there remains a pressing need to develop a reliable tool that integrates multimodal data—such as clinical features, cognitive functions, electroencephalographic microstates, and inflammation-associated proteins—to improve diagnostic accuracy and objectivity. One hundred and fifteen patients with depression and 66 healthy controls were included in this study, and data on their clinical characteristics, cognitive function, electroencephalographic microstates, and serum inflammation-related proteins were collected. The baseline depression group was followed up after 4 weeks of clinical treatment, with 56 participants completing the follow-up survey, which mirrored the baseline survey. The depression baseline and healthy control groups were designated as the training set, while the follow-up and healthy control groups served as the validation set. Six classical machine learning algorithms—Decision Tree, Random Forest, XGBoost, LightGBM, k-Nearest Neighbor, and Support Vector Machine—were employed to train the diagnostic prediction model using the training set. The model was then validated with the validation set to identify the optimal depression diagnostic prediction model. The results of the study showed significant differences in clinical characteristics, cognitive function, EEG microstates, and serum levels of inflammation-related proteins in patients with major depressive disorder compared with healthy controls. In the model evaluation, the k-nearest neighbor model performed the best, with an accuracy of 95.08% for multimodal diagnosis, an F1 score of 0.9545, and an AUC value of 0.9969. In order of feature importance were IL-8, IL-18, ISI, MMP-8, CD40, CASP-8, visuospatial/constructional, and mean duration of EEG microstate D. A multimodal, multi-indicator model (incorporating L-8, IL-18, ISI, MMP-8, CD40, CASP-8, visuospatial/constructional abilities, and mean duration of EEG microstate D) has the potential to enhance the accuracy and objectivity of clinical depression diagnoses.