Improved ASD classification using dynamic functional connectivity and multi-task feature selection

• Published in: Pattern recognition letters Vol. 138; pp. 82 - 87
• Main Authors: Liu, Jin, Sheng, Yu, Lan, Wei, Guo, Rui, Wang, Yufei, Wang, Jianxin
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 01.10.2020; Elsevier Science Ltd
• Subjects: ASD Classification; Autism; Brain mapping; Classification; Clinical medicine; Communication skills; Data exchange; Diagnosis; Dynamic functional connectivity; Feature extraction; Functional magnetic resonance imaging; Magnetic resonance imaging; Multi-kernel learning; Multi-task feature selection; Neural networks; Neurodevelopmental disorders; Neuroimaging; Resting state functional MRI; Multi-task feature selection; Resting state functional MRI; 41A10; 65D05; 65D17; Dynamic functional connectivity; Multi-kernel learning; 41A05; ASD Classification
• ISSN: 0167-8655; 1872-7344
• DOI: 10.1016/j.patrec.2020.07.005

•We propose an efficient method for diagnosing ASD.•Dynamic functional connectivity features have complementary information.•We proposed an improved multi-task feature selection method.•A multi-kernel SVM method is applied.•Our method achieves classification accuracy of 76.8%. Accurate diagnosis of autism spectrum disorder (ASD), which is a neurodevelopmental disorder and often accompanied by abnormal social skills, communication skills, interests and behavior patterns, has always been a challenging task in clinical practice. Recent studies have shown great potential for using fMRI data to distinguish ASD from typical control (TC). However, it has always been a challenging problem to extract which features from fMRI data and how to combine these different types of features to achieve improved ASD/TC classification performance. To address this problem, in this study we propose an improved ASD/TC classification framework based on dynamic functional connectivity (DFC) and multi-task feature selection. Our proposed ASD/TC classification framework is evaluated on 871 subjects with fMRI data from the Autism Brain Imaging Data Exchange I (ABIDE I) via a 10-fold cross validation strategy. Experimental results show that our proposed method achieves an accuracy of 76.8% and an area under the receiver operating characteristic curve (AUC) of 0.81 for ASD/TC classification. In addition, compared with some existing state-of-the-art methods, our proposed method achieves better accuracy and AUC for ASD/TC classification. Overall, our proposed ASD/TC classification framework is effective and promising for automatic diagnosis of ASD in clinical practice.