相互信息
认知
人工智能
功能磁共振成像
计算机科学
模式识别(心理学)
特征选择
特征(语言学)
相关性
认知网络
心理学
数学
神经科学
认知无线电
电信
语言学
哲学
几何学
无线
作者
Xiaofei Zhang,Yang Yang,Ning Zhong
标识
DOI:10.1109/wi-iat59888.2023.00086
摘要
The pre-defined regions of interest (ROIs) in brain atlases can be employed as features for classifying mental states during the cognitive procedure. The ROIs that are useful for identifying mental states can be thought of as related features, or cognitive features, that point to the specific cognitive function in the human brain. A mutual information based cognitive feature selection method (MI-CFSM) was proposed to resolve the ambiguity of cognitive feature selection in brain atlas. Firstly, the blood oxygenation level dependent (BOLD) signals in the functional magnetic resonance imaging (fMRI) data were extracted based on the pre-defined ROIs. Secondly, the mutual information between cognitive features and cognitive states were calculated, and the cognitive features were ranked based on the calculated mutual information. Finally, the cognitive network architecture - area under the curve (CNA-AUC) values for the ranked cognitive features were determined, and the performances of numerous cognitive feature selection approaches were assessed. In the experiment conducted on fMRI data of mental arithmetic cognitive tasks, the CNA-AUC values obtained by MI-CFSM on the task positive correlation system (TPS), task negative correlation system (TNS), and task support system (TSS) of the cognitive network architecture were 0.5929, 0.4704, and 0.4464, respectively. Compared with the other adopted methods, the MI-CFSM method has the highest TPS/TNS ratio, although the CNA-AUC of TPS is not the highest. This method generally tends to choose the ROIs belonging to the former as the cognitive features between TPS and TNS, which better reflects the category and function of the cognitive network architecture.
科研通智能强力驱动
Strongly Powered by AbleSci AI