BrainOSM: Outlier screening for multi-view functional brain network analysis.

Area of Science:

• Neuroscience
• Computational Biology
• Medical Informatics

Background:

• Identifying reliable biomarkers for mental diseases is crucial for early diagnosis and personalized treatment.
• Functional brain networks (FBNs), represented as graphs, capture brain connectivity but face challenges like heterogeneity and noise.
• Autism spectrum disorder (ASD) and Alzheimer's disease (AD) diagnosis requires robust methods for analyzing complex FBNs.

Purpose of the Study:

• To develop and validate a novel framework, BrainOSM, for diagnosing ASD and AD using FBNs.
• To address the challenges of graph heterogeneity and disease-unrelated information in FBN analysis.
• To improve the accuracy and generalizability of mental disease classification through advanced graph analysis.

Main Methods:

• Introduced BrainOSM, a two-stage framework combining outlier screening and multi-view graph pooling.
• Employed progressive uncertainty-based outlier screening to mitigate inter-graph heterogeneity.
• Integrated multi-graph pooling, multi-view learning, and prior subnetwork regularization to refine graph structures and reduce noise.

Main Results:

• BrainOSM demonstrated superior performance on the ABIDE (ASD) and ADNI (AD) datasets compared to traditional Graph Convolutional Network (GCN) methods.
• Achieved an average accuracy of 70.23% and AUC of 70.42% on ABIDE, outperforming GCN by 8.55% and 7.74%.
• Reached an average accuracy of 82.29% and AUC of 83.23% on ADNI, with improvements of 8.97% and 11.78% over GCN.

Conclusions:

• BrainOSM is a generalizable and effective framework for mental disease classification using FBNs.
• The method successfully identifies disease-associated subnetworks, offering potential for clinical interpretation.
• Outlier screening is a critical component for enhancing classification accuracy in heterogeneous neuroimaging datasets.