Scholarlink

Purpose <p>Attention-deficit/hyperactivity disorder (ADHD) is a chronic condition that affects about 5% of the children. This research presents a novel method to identify ADHD and its subtype biomarkers using frequency band electroencephalography (EEG) data through the fusion of random forest (RF) models.</p> Methods <p>This work used EEG data from 776  participants from the Healthy Brain Network initiative database (collected by the Child Mind Institute). After performing quality control, the number of available subjects is reduced to 149 controls, 169 with combined ADHD, and 181 with inattentive ADHD. Data was carefully preprocessed. RF classifier optimized through genetic algorithms then classified the principal component analysis (PCA) selected features. The fusion stage uses RF predictions for each frequency band as feature.</p> Results <p>Among the tested fusion methods, we highlight the ridge linear model and XGBoost, which achieved 92.1 ± 0.5% and 97.7 ± 0.3% accuracies in ADHD/control and subtype classifications, respectively. Most optimization methods pointed out the greater importance for the Delta band, and biomarker analysis indicates that the information from different frequency bands is complementary. The study also pointed out the importance of Alpha and Gamma for detecting ADHD and of Theta for subtype diagnosis.</p> Conclusion <p>The present research may also be a base for further studies to develop new treatments.</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

Investigation of electroencephalography in attention-deficit hyperactivity disorder subtype classification with machine learning

  • 期刊论文

  • 发表日期：

    2025年2月28日

查看全文
  • Guilherme Rodrigues Pedrollo,
  • Alexandre Rosa Franco,
  • Alexandre Balbinot

摘要

Purpose

Attention-deficit/hyperactivity disorder (ADHD) is a chronic condition that affects about 5% of the children. This research presents a novel method to identify ADHD and its subtype biomarkers using frequency band electroencephalography (EEG) data through the fusion of random forest (RF) models.

Methods

This work used EEG data from 776  participants from the Healthy Brain Network initiative database (collected by the Child Mind Institute). After performing quality control, the number of available subjects is reduced to 149 controls, 169 with combined ADHD, and 181 with inattentive ADHD. Data was carefully preprocessed. RF classifier optimized through genetic algorithms then classified the principal component analysis (PCA) selected features. The fusion stage uses RF predictions for each frequency band as feature.

Results

Among the tested fusion methods, we highlight the ridge linear model and XGBoost, which achieved 92.1 ± 0.5% and 97.7 ± 0.3% accuracies in ADHD/control and subtype classifications, respectively. Most optimization methods pointed out the greater importance for the Delta band, and biomarker analysis indicates that the information from different frequency bands is complementary. The study also pointed out the importance of Alpha and Gamma for detecting ADHD and of Theta for subtype diagnosis.

Conclusion

The present research may also be a base for further studies to develop new treatments.