面向癫痫的脑电图可视分析方法
An Electroencephalogram Visual Analytics Approach for Epilepsy
丁国明 1谢潇 2胡玲利 3巫英才 1王爽3
作者信息
- 1. 浙江大学CAD&CG国家重点实验室 杭州 310058
- 2. 浙江大学教育学院体育学系 杭州 310058
- 3. 浙江大学医学院附属第二医院 杭州 310058
- 折叠
摘要
为了高效地检测和分析癫痫患者脑电图中的癫痫样放电,提出一种面向癫痫的脑电图可视分析方法.首先利用浙江大学医学院附属第二医院在癫痫治疗过程中积累的脑电数据,建立基于时域和时频域视角的卷积神经网络模型,用于癫痫样放电自动分类,使用Focal Loss作为损失函数评估放电类别分布不均衡的问题;然后在分类模型基础上,结合医生的需求设计交互式的多视图可视分析系统,其由波形分类视图、离群值检测视图和波形检查视图3个相互协作的模块组成;最后招募10位被试和2位专家,通过评估实验验证系统在癫痫诊断方面的有效性和可用性.实验结果表明,建立的分类模型在所用数据集上的平均F1分数为0.88,优于其他的脑电分类模型,且具有良好的鲁棒性;设计的系统具有良好的可用性和有效性,可以辅助医生进行癫痫诊断.
Abstract
In order to efficiently detect and analyze epileptiform discharges in electroencephalogram of epi-lepsy patients,an electroencephalogram visual analytics approach for epilepsy is proposed.Firstly,using the electroencephalogram data accumulated by the Second Affiliated Hospital of Zhejiang University School of Medicine during epilepsy treatment,a convolutional neural network model based on the perspective of time domain and time-frequency domain is established for the automatic classification of epileptiform discharges,and Focal Loss is used as the loss function to evaluate the unbalanced distribution of discharge categories.Secondly,based on the classification model and combined with the needs of doctors,an interactive multi-view visual analysis system is designed,which consists of three cooperative modules:waveform clas-sification view,outlier detection view and waveform inspection view.Then,10 participants and 2 experts were recruited to verify the effectiveness and usability of the system in epilepsy diagnosis by evaluating ex-periments.The results show that the average F1 score of the established classification model on the dataset used is 0.88,which is better than other classification models and has good robustness.The system is de-signed with good usability and effectiveness,which can assist doctors in epilepsy diagnosis.
关键词
医学可视化/机器学习/脑电图/癫痫Key words
medical visualization/machine learning/electroencephalogram/epilepsy引用本文复制引用
出版年
2024
NETL
NSTL
万方数据