静息态功能连接特异性与机器学习的癫痫定侧
COMBINING RESTING-STATE FUNCTIONAL CONNECTIVITY SPECIFICITY AND MACHINE LEARNING TO LOCALIZE PAROXYSMAL SIDE OF EPILEPTIC PATIENTS
宋子博 1葛曼玲 1付晓璇 1陈盛华 1郭志彤 1张其锐 2张志强2
作者信息
- 1. 河北工业大学省部共建电工装备可靠性与智能化国家重点实验室 天津 300130;河北工业大学河北省电磁场与电器可靠性重点实验室 天津 300130
- 2. 南京大学医学院附属金陵医院/东部战区总医院医学影像科 江苏南京 210002
- 折叠
摘要
为探索癫痫发作侧的脑功能影像标记,提出静息态功能磁共振的功能连接特异性模型和有监督机器学习联合方案.选取20名结构影像提示发作侧的颞叶癫痫患者(均分左、右两组)和142名健康人;以健康人为参照,构建功能连接特异性模型,为每位患者每个脑区功能连接打分;统计分析左右组间打分值差异显著性,获得对发作侧敏感的标志性脑区;以其打分值为特征向量输入到概率神经网络实现定侧并使用交叉验证.结果显示,对发作侧敏感的功能影像学标记在杏仁核、中央旁小叶等6个脑区,分类准确率达90.0％,高于目前机器学习辅助癫痫研究水准.
Abstract
To explore the functional brain imaging markers of epileptic seizure side,a joint scheme of functional connectivity specificity modeling and supervised machine learning with resting-state functional magnetic resonance is proposed.Twenty temporal lobe epilepsy patients with structural images suggestive of the seizure side(equally divided into left and right groups)and 142 healthy individuals were selected.We used healthy individuals as reference,and a functional connectivity specificity model was constructed to score the functional connectivity of each brain region for each patient.The significance of the difference in scoring values between the left and right groups was statistically analyzed to obtain the landmark brain regions that were sensitive to the seizure side.The scoring values were used as a feature vector inputted into a probabilistic neural network to achieve the fixation of the side and cross validation was used.The results show that:functional imaging markers sensitive to the ictal side are in six brain regions,including the amygdala and paracentral lobule,with a classification accuracy of 90.0％,which is higher than the current level of machine learning-assisted epilepsy research.
关键词
静息态功能磁共振/功能连接特异性/概率神经网络/颞叶癫痫/发作侧Key words
Resting-state functional magnetic resonance/Functional connectivity specificity/Probabilistic neural network/Temporal lobe epilepsy/Seizure lateralization引用本文复制引用
基金项目
国家自然科学基金项目(81871345)
河北省自然科学基金项目(E2019202019)
出版年
2024
国家科技期刊平台
NSTL
万方数据